# Research Papers — 2026-01 · bundle page 22 of 189

> Full-text markdown of papers #1051–1100 of 9403 published in 2026-01, newest first. AI-readable mirrors of the full 9403-paper 2026-01 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-01/page/22
[← Previous bundle](https://aipod.dev/papers/archive/2026-01/page/21.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-01/page/23.md)

### Prior-Informed Flow Matching for Graph Reconstruction

Authors: Harvey Chen, Nicolas Zilberstein, Santiago Segarra
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22107v2

We introduce \textit{Prior-Informed Flow Matching (PIFM)}, a conditional flow model for graph reconstruction. Reconstructing graphs from partial observations remains a key challenge; classical embedding methods often lack global consistency, while modern generative models struggle to incorporate structural priors. PIFM bridges this gap by integrating embedding-based priors with continuous-time flow matching. Grounded in a permutation equivariant version of the distortion-perception theory, our method first uses a prior, such as GraphSAGE or node2vec, to form an informed initial estimate of the adjacency matrix based on local information. It then applies rectified flow matching to refine this estimate, transporting it toward the true distribution of clean graphs and learning a global coupling. Experiments on different datasets demonstrate that PIFM consistently enhances classical embeddings, outperforming them and state-of-the-art generative baselines in reconstruction accuracy.

---

### ECO: Quantized Training without Full-Precision Master Weights

Authors: Mahdi Nikdan, Amir Zandieh, Dan Alistarh, Vahab Mirrokni
Published: 2026-01-29
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.22101v1

Quantization has significantly improved the compute and memory efficiency of Large Language Model (LLM) training. However, existing approaches still rely on accumulating their updates in high-precision: concretely, gradient updates must be applied to a high-precision weight buffer, known as $\textit{master weights}$. This buffer introduces substantial memory overhead, particularly for Sparse Mixture of Experts (SMoE) models, where model parameters and optimizer states dominate memory usage. To address this, we introduce the Error-Compensating Optimizer (ECO), which eliminates master weights by applying updates directly to quantized parameters. ECO quantizes weights after each step and carefully injects the resulting quantization error into the optimizer momentum, forming an error-feedback loop with no additional memory. We prove that, under standard assumptions and a decaying learning rate, ECO converges to a constant-radius neighborhood of the optimum, while naive master-weight removal can incur an error that is inversely proportional to the learning rate. We show empirical results for pretraining small Transformers (30-800M), a Gemma-3 1B model, and a 2.1B parameter Sparse MoE model with FP8 quantization, and fine-tuning DeepSeek-MoE-16B in INT4 precision. Throughout, ECO matches baselines with master weights up to near-lossless accuracy, significantly shifting the static memory vs validation loss Pareto frontier.

---

### Boosting CVaR Policy Optimization with Quantile Gradients

Authors: Yudong Luo, Erick Delage
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22100v3

Optimizing Conditional Value-at-risk (CVaR) using policy gradient (a.k.a CVaR-PG) faces significant challenges of sample inefficiency. This inefficiency stems from the fact that it focuses on tail-end performance and overlooks many sampled trajectories. We address this problem by augmenting CVaR with an expected quantile term. Quantile optimization admits a dynamic programming formulation that leverages all sampled data, thus improves sample efficiency. This does not alter the CVaR objective since CVaR corresponds to the expectation of quantile over the tail. Empirical results in domains with verifiable risk-averse behavior show that our algorithm within the Markovian policy class substantially improves upon CVaR-PG and consistently outperforms other existing methods.

---

### GeoNorm: Unify Pre-Norm and Post-Norm with Geodesic Optimization

Authors: Chuanyang Zheng, Jiankai Sun, Yihang Gao, Chi Wang, Yuehao Wang, Jing Xiong, Liliang Ren, Bo Peng, Qingmei Wang, Xiaoran Shang, Mac Schwager, Anderson Schneider et al.
Published: 2026-01-29
Categories: cs.LG, cs.CL
arXiv: https://arxiv.org/abs/2601.22095v1

The placement of normalization layers, specifically Pre-Norm and Post-Norm, remains an open question in Transformer architecture design. In this work, we rethink these approaches through the lens of manifold optimization, interpreting the outputs of the Feed-Forward Network (FFN) and attention layers as update directions in optimization. Building on this perspective, we introduce GeoNorm, a novel method that replaces standard normalization with geodesic updates on the manifold. Furthermore, analogous to learning rate schedules, we propose a layer-wise update decay for the FFN and attention components. Comprehensive experiments demonstrate that GeoNorm consistently outperforms existing normalization methods in Transformer models. Crucially, GeoNorm can be seamlessly integrated into standard Transformer architectures, achieving performance improvements with negligible additional computational cost.

---

### RefAny3D: 3D Asset-Referenced Diffusion Models for Image Generation

Authors: Hanzhuo Huang, Qingyang Bao, Zekai Gu, Zhongshuo Du, Cheng Lin, Yuan Liu, Sibei Yang
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.22094v1

In this paper, we propose a 3D asset-referenced diffusion model for image generation, exploring how to integrate 3D assets into image diffusion models. Existing reference-based image generation methods leverage large-scale pretrained diffusion models and demonstrate strong capability in generating diverse images conditioned on a single reference image. However, these methods are limited to single-image references and cannot leverage 3D assets, constraining their practical versatility. To address this gap, we present a cross-domain diffusion model with dual-branch perception that leverages multi-view RGB images and point maps of 3D assets to jointly model their colors and canonical-space coordinates, achieving precise consistency between generated images and the 3D references. Our spatially aligned dual-branch generation architecture and domain-decoupled generation mechanism ensure the simultaneous generation of two spatially aligned but content-disentangled outputs, RGB images and point maps, linking 2D image attributes with 3D asset attributes. Experiments show that our approach effectively uses 3D assets as references to produce images consistent with the given assets, opening new possibilities for combining diffusion models with 3D content creation.

---

### Investigating Associational Biases in Inter-Model Communication of Large Generative Models

Authors: Fethiye Irmak Dogan, Yuval Weiss, Kajal Patel, Jiaee Cheong, Hatice Gunes
Published: 2026-01-29
Categories: cs.CY, cs.AI
arXiv: https://arxiv.org/abs/2601.22093v1

Social bias in generative AI can manifest not only as performance disparities but also as associational bias, whereby models learn and reproduce stereotypical associations between concepts and demographic groups, even in the absence of explicit demographic information (e.g., associating doctors with men). These associations can persist, propagate, and potentially amplify across repeated exchanges in inter-model communication pipelines, where one generative model's output becomes another's input. This is especially salient for human-centred perception tasks, such as human activity recognition and affect prediction, where inferences about behaviour and internal states can lead to errors or stereotypical associations that propagate into unequal treatment. In this work, focusing on human activity and affective expression, we study how such associations evolve within an inter-model communication pipeline that alternates between image generation and image description. Using the RAF-DB and PHASE datasets, we quantify demographic distribution drift induced by model-to-model information exchange and assess whether these drifts are systematic using an explainability pipeline. Our results reveal demographic drifts toward younger representations for both actions and emotions, as well as toward more female-presenting representations, primarily for emotions. We further find evidence that some predictions are supported by spurious visual regions (e.g., background or hair) rather than concept-relevant cues (e.g., body or face). We also examine whether these demographic drifts translate into measurable differences in downstream behaviour, i.e., while predicting activity and emotion labels. Finally, we outline mitigation strategies spanning data-centric, training and deployment interventions, and emphasise the need for careful safeguards when deploying interconnected models in human-centred AI systems.

---

### Learning Transient Convective Heat Transfer with Geometry Aware World Models

Authors: Onur T. Doganay, Alexander Klawonn, Martin Eigel, Hanno Gottschalk
Published: 2026-01-29
Categories: physics.flu-dyn, cs.CV
arXiv: https://arxiv.org/abs/2601.22086v1

Partial differential equation (PDE) simulations are fundamental to engineering and physics but are often computationally prohibitive for real-time applications. While generative AI offers a promising avenue for surrogate modeling, standard video generation architectures lack the specific control and data compatibility required for physical simulations. This paper introduces a geometry aware world model architecture, derived from a video generation architecture (LongVideoGAN), designed to learn transient physics. We introduce two key architecture elements: (1) a twofold conditioning mechanism incorporating global physical parameters and local geometric masks, and (2) an architectural adaptation to support arbitrary channel dimensions, moving beyond standard RGB constraints. We evaluate this approach on a 2D transient computational fluid dynamics (CFD) problem involving convective heat transfer from buoyancy-driven flow coupled to a heat flow in a solid structure. We demonstrate that the conditioned model successfully reproduces complex temporal dynamics and spatial correlations of the training data. Furthermore, we assess the model's generalization capabilities on unseen geometric configurations, highlighting both its potential for controlled simulation synthesis and current limitations in spatial precision for out-of-distribution samples.

---

### Latent Adversarial Regularization for Offline Preference Optimization

Authors: Enyi Jiang, Yibo Jacky Zhang, Yinglun Xu, Andreas Haupt, Nancy Amato, Sanmi Koyejo
Published: 2026-01-29
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.22083v2

Learning from human feedback typically relies on preference optimization that constrains policy updates through token-level regularization. However, preference optimization for language models is particularly challenging because token-space similarity does not imply semantic or behavioral similarity. To address this challenge, we leverage latent-space regularization for language model preference optimization. We introduce GANPO, which achieves latent-space regularization by penalizing divergence between the internal representations of a policy model and a reference model. Given that latent representations are not associated with explicit probability densities, we adopt an adversarial approach inspired by GANs to minimize latent-space divergence. We integrate GANPO as a regularizer into existing offline preference optimization objectives. Experiments across multiple model architectures and tasks show consistent improvements from latent-space regularization. Further, by comparing GANPO-induced inferential biases with those from token-level regularization, we find that GANPO provides more robust structural feedback under distributional shift and noise while maintaining comparable downstream performance with minor computational overhead.

---

### Where Do the Joules Go? Diagnosing Inference Energy Consumption

Authors: Jae-Won Chung, Ruofan Wu, Jeff J. Ma, Mosharaf Chowdhury
Published: 2026-01-29
Categories: cs.LG, cs.DC
arXiv: https://arxiv.org/abs/2601.22076v2

Energy is now a critical ML computing resource. While measuring energy consumption and observing trends is a valuable first step, accurately understanding and diagnosing why those differences occur is crucial for optimization. To that end, we begin by presenting a large-scale measurement study of inference time and energy across the generative AI landscape with 46 models, 7 tasks, and 1,858 different configurations on NVIDIA H100 and B200 GPUs. Our empirical findings span order-of-magnitude variations: LLM task type can lead to 25$\times$ energy differences, video generation sometimes consumes more than 100$\times$ the energy of images, and GPU utilization differences can result in 3--5$\times$ energy differences. Based on our observations, we present a framework for reasoning about the underlying mechanisms that govern time and energy consumption. The essence is that time and energy are determined by latent metrics like memory and utilization, which are in turn affected by various factors across the algorithm, software, and hardware layers. Our framework also extends directly to throughput per watt, a critical metric for power-constrained datacenters.

---

### Lens-descriptor guided evolutionary algorithm for optimization of complex optical systems with glass choice

Authors: Kirill Antonov, Teus Tukker, Tiago Botari, Thomas H. W. Bäck, Anna V. Kononova, Niki van Stein
Published: 2026-01-29
Categories: cs.NE
arXiv: https://arxiv.org/abs/2601.22075v1

Designing high-performance optical lenses entails exploring a high-dimensional, tightly constrained space of surface curvatures, glass choices, element thicknesses, and spacings. In practice, standard optimizers (e.g., gradient-based local search and evolutionary strategies) often converge to a single local optimum, overlooking many comparably good alternatives that matter for downstream engineering decisions. We propose the Lens Descriptor-Guided Evolutionary Algorithm (LDG-EA), a two-stage framework for multimodal lens optimization. LDG-EA first partitions the design space into behavior descriptors defined by curvature-sign patterns and material indices, then learns a probabilistic model over descriptors to allocate evaluations toward promising regions. Within each descriptor, LDG-EA applies the Hill-Valley Evolutionary Algorithm with covariance-matrix self-adaptation to recover multiple distinct local minima, optionally followed by gradient-based refinement. On a 24-variable (18 continuous and 6 integer), six-element Double-Gauss topology, LDG-EA generates on average around 14500 candidate minima spanning 636 unique descriptors, an order of magnitude more than a CMA-ES baseline, while keeping wall-clock time at one hour scale. Although the best LDG-EA design is slightly worse than a fine-tuned reference lens, it remains in the same performance range. Overall, the proposed LDG-EA produces a diverse set of solutions while maintaining competitive quality within practical computational budgets and wall-clock time.

---

### VTC-R1: Vision-Text Compression for Efficient Long-Context Reasoning

Authors: Yibo Wang, Yongcheng Jing, Shunyu Liu, Hao Guan, Rong-cheng Tu, Chengyu Wang, Jun Huang, Dacheng Tao
Published: 2026-01-29
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.22069v2

Long-context reasoning has significantly empowered large language models (LLMs) to tackle complex tasks, yet it introduces severe efficiency bottlenecks due to the computational complexity. Existing efficient approaches often rely on complex additional training or external models for compression, which limits scalability and discards critical fine-grained information. In this paper, we propose VTC-R1, a new efficient reasoning paradigm that integrates vision-text compression into the reasoning process. Instead of processing lengthy textual traces, VTC-R1 renders intermediate reasoning segments into compact images, which are iteratively fed back into vision-language models as "optical memory." We construct a training dataset based on OpenR1-Math-220K achieving 3.4x token compression and fine-tune representative VLMs-Glyph and Qwen3-VL. Extensive experiments on benchmarks such as MATH500, AIME25, AMC23 and GPQA-D demonstrate that VTC-R1 consistently outperforms standard long-context reasoning. Furthermore, our approach significantly improves inference efficiency, achieving 2.7x speedup in end-to-end latency, highlighting its potential as a scalable solution for reasoning-intensive applications. Our code is available at https://github.com/w-yibo/VTC-R1.

---

### Quantifying the Uncertainty of Foundation Models with Singular Value Ensembles

Authors: Mehmet Ozgur Turkoglu, Dominik J. Mühlematter, Alexander Becker, Konrad Schindler, Helge Aasen
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22068v2

Foundation models have become a dominant paradigm in machine learning, achieving remarkable performance across diverse tasks through large-scale pretraining. However, they often yield overconfident, uncalibrated predictions. The standard approach to quantifying epistemic uncertainty are ensembles of multiple independently trained models. But their computational cost scales linearly with ensemble size, making them impractical for large foundation models. We propose Singular Value Ensemble (SVE), a parameter-efficient implicit ensembling method. SVE builds on a simple, but powerful core assumption: namely, that the singular vectors of the weight matrices correspond to meaningful directions in the representation space. If the singular vectors are indeed meaningful (orthogonal) "knowledge directions", then a model ensemble can be obtained by modulating only how strongly each direction contributes to the output. Rather than learning new parameters for each ensemble member, we freeze the singular vectors and only train per-member singular values that rescale the contribution of each direction in that shared knowledge basis. Ensemble diversity emerges naturally during joint training as stochastic initialization and random batch sampling cause different members to converge to different combinations of the same underlying knowledge. SVE performs comparable to an explicit ensemble, while increasing the parameter count of the base model by <1%, making principled uncertainty estimation accessible in resource-constrained settings. We validate SVE on NLP and vision tasks with various different backbones and show that it improves calibration while maintaining predictive accuracy.

---

### BLO-Inst: Bi-Level Optimization Based Alignment of YOLO and SAM for Robust Instance Segmentation

Authors: Li Zhang, Pengtao Xie
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.22061v1

The Segment Anything Model has revolutionized image segmentation with its zero-shot capabilities, yet its reliance on manual prompts hinders fully automated deployment. While integrating object detectors as prompt generators offers a pathway to automation, existing pipelines suffer from two fundamental limitations: objective mismatch, where detectors optimized for geometric localization do not correspond to the optimal prompting context required by SAM, and alignment overfitting in standard joint training, where the detector simply memorizes specific prompt adjustments for training samples rather than learning a generalizable policy. To bridge this gap, we introduce BLO-Inst, a unified framework that aligns detection and segmentation objectives by bi-level optimization. We formulate the alignment as a nested optimization problem over disjoint data splits. In the lower level, the SAM is fine-tuned to maximize segmentation fidelity given the current detection proposals on a subset ($D_1$). In the upper level, the detector is updated to generate bounding boxes that explicitly minimize the validation loss of the fine-tuned SAM on a separate subset ($D_2$). This effectively transforms the detector into a segmentation-aware prompt generator, optimizing the bounding boxes not just for localization accuracy, but for downstream mask quality. Extensive experiments demonstrate that BLO-Inst achieves superior performance, outperforming standard baselines on tasks in general and biomedical domains.

---

### Vision-DeepResearch: Incentivizing DeepResearch Capability in Multimodal Large Language Models

Authors: Wenxuan Huang, Yu Zeng, Qiuchen Wang, Zhen Fang, Shaosheng Cao, Zheng Chu, Qingyu Yin, Shuang Chen, Zhenfei Yin, Lin Chen, Zehui Chen, Xu Tang et al.
Published: 2026-01-29
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.22060v3

Multimodal large language models (MLLMs) have achieved remarkable success across a broad range of vision tasks. However, constrained by the capacity of their internal world knowledge, prior work has proposed augmenting MLLMs by ``reasoning-then-tool-call'' for visual and textual search engines to obtain substantial gains on tasks requiring extensive factual information. However, these approaches typically define multimodal search in a naive setting, assuming that a single full-level or entity-level image query and few text query suffices to retrieve the key evidence needed to answer the question, which is unrealistic in real-world scenarios with substantial visual noise. Moreover, they are often limited in the reasoning depth and search breadth, making it difficult to solve complex questions that require aggregating evidence from diverse visual and textual sources. Building on this, we propose Vision-DeepResearch, which proposes one new multimodal deep-research paradigm, i.e., performs multi-turn, multi-entity and multi-scale visual and textual search to robustly hit real-world search engines under heavy noise. Our Vision-DeepResearch supports dozens of reasoning steps and hundreds of engine interactions, while internalizing deep-research capabilities into the MLLM via cold-start supervision and RL training, resulting in a strong end-to-end multimodal deep-research MLLM. It substantially outperforming existing multimodal deep-research MLLMs, and workflows built on strong closed-source foundation model such as GPT-5, Gemini-2.5-pro and Claude-4-Sonnet. The code will be released in https://github.com/Osilly/Vision-DeepResearch.

---

### Unsupervised Decomposition and Recombination with Discriminator-Driven Diffusion Models

Authors: Archer Wang, Emile Anand, Yilun Du, Marin Soljačić
Published: 2026-01-29
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.22057v2

Decomposing complex data into factorized representations can reveal reusable components and enable synthesizing new samples via component recombination. We investigate this in the context of diffusion-based models that learn factorized latent spaces without factor-level supervision. In images, factors can capture background, illumination, and object attributes; in robotic videos, they can capture reusable motion components. To improve both latent factor discovery and quality of compositional generation, we introduce an adversarial training signal via a discriminator trained to distinguish between single-source samples and those generated by recombining factors across sources. By optimizing the generator to fool this discriminator, we encourage physical and semantic consistency in the resulting recombinations. Our method outperforms implementations of prior baselines on CelebA-HQ, Virtual KITTI, CLEVR, and Falcor3D, achieving lower FID scores and better disentanglement as measured by MIG and MCC. Furthermore, we demonstrate a novel application to robotic video trajectories: by recombining learned action components, we generate diverse sequences that significantly increase state-space coverage for exploration on the LIBERO benchmark.

---

### $G^2$-Reader: Dual Evolving Graphs for Multimodal Document QA

Authors: Yaxin Du, Junru Song, Yifan Zhou, Cheng Wang, Jiahao Gu, Zimeng Chen, Menglan Chen, Wen Yao, Yang Yang, Ying Wen, Siheng Chen
Published: 2026-01-29
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.22055v1

Retrieval-augmented generation is a practical paradigm for question answering over long documents, but it remains brittle for multimodal reading where text, tables, and figures are interleaved across many pages. First, flat chunking breaks document-native structure and cross-modal alignment, yielding semantic fragments that are hard to interpret in isolation. Second, even iterative retrieval can fail in long contexts by looping on partial evidence or drifting into irrelevant sections as noise accumulates, since each step is guided only by the current snippet without a persistent global search state. We introduce $G^2$-Reader, a dual-graph system, to address both issues. It evolves a Content Graph to preserve document-native structure and cross-modal semantics, and maintains a Planning Graph, an agentic directed acyclic graph of sub-questions, to track intermediate findings and guide stepwise navigation for evidence completion. On VisDoMBench across five multimodal domains, $G^2$-Reader with Qwen3-VL-32B-Instruct reaches 66.21\% average accuracy, outperforming strong baselines and a standalone GPT-5 (53.08\%).

---

### MetricAnything: Scaling Metric Depth Pretraining with Noisy Heterogeneous Sources

Authors: Baorui Ma, Jiahui Yang, Donglin Di, Xuancheng Zhang, Jianxun Cui, Hao Li, Yan Xie, Wei Chen
Published: 2026-01-29
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.22054v2

Scaling has powered recent advances in vision foundation models, yet extending this paradigm to metric depth estimation remains challenging due to heterogeneous sensor noise, camera-dependent biases, and metric ambiguity in noisy cross-source 3D data. We introduce Metric Anything, a simple and scalable pretraining framework that learns metric depth from noisy, diverse 3D sources without manually engineered prompts, camera-specific modeling, or task-specific architectures. Central to our approach is the Sparse Metric Prompt, created by randomly masking depth maps, which serves as a universal interface that decouples spatial reasoning from sensor and camera biases. Using about 20M image-depth pairs spanning reconstructed, captured, and rendered 3D data across 10000 camera models, we demonstrate-for the first time-a clear scaling trend in the metric depth track. The pretrained model excels at prompt-driven tasks such as depth completion, super-resolution and Radar-camera fusion, while its distilled prompt-free student achieves state-of-the-art results on monocular depth estimation, camera intrinsics recovery, single/multi-view metric 3D reconstruction, and VLA planning. We also show that using pretrained ViT of Metric Anything as a visual encoder significantly boosts Multimodal Large Language Model capabilities in spatial intelligence. These results show that metric depth estimation can benefit from the same scaling laws that drive modern foundation models, establishing a new path toward scalable and efficient real-world metric perception. We open-source MetricAnything at http://metric-anything.github.io/metric-anything-io/ to support community research.

---

### MasalBench: A Benchmark for Contextual and Cross-Cultural Understanding of Persian Proverbs in LLMs

Authors: Ghazal Kalhor, Behnam Bahrak
Published: 2026-01-29
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.22050v1

In recent years, multilingual Large Language Models (LLMs) have become an inseparable part of daily life, making it crucial for them to master the rules of conversational language in order to communicate effectively with users. While previous work has evaluated LLMs' understanding of figurative language in high-resource languages, their performance in low-resource languages remains underexplored. In this paper, we introduce MasalBench, a comprehensive benchmark for assessing LLMs' contextual and cross-cultural understanding of Persian proverbs, which are a key component of conversation in this low-resource language. We evaluate eight state-of-the-art LLMs on MasalBench and find that they perform well in identifying Persian proverbs in context, achieving accuracies above 0.90. However, their performance drops considerably when tasked with identifying equivalent English proverbs, with the best model achieving 0.79 accuracy. Our findings highlight the limitations of current LLMs in cultural knowledge and analogical reasoning, and they provide a framework for assessing cross-cultural understanding in other low-resource languages. MasalBench is available at https://github.com/kalhorghazal/MasalBench.

---

### On the Paradoxical Interference between Instruction-Following and Task Solving

Authors: Yunjia Qi, Hao Peng, Xintong Shi, Amy Xin, Xiaozhi Wang, Bin Xu, Lei Hou, Juanzi Li
Published: 2026-01-29
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.22047v1

Instruction following aims to align Large Language Models (LLMs) with human intent by specifying explicit constraints on how tasks should be performed. However, we reveal a counterintuitive phenomenon: instruction following can paradoxically interfere with LLMs' task-solving capability. We propose a metric, SUSTAINSCORE, to quantify the interference of instruction following with task solving. It measures task performance drop after inserting into the instruction a self-evident constraint, which is naturally met by the original successful model output and extracted from it. Experiments on current LLMs in mathematics, multi-hop QA, and code generation show that adding the self-evident constraints leads to substantial performance drops, even for advanced models such as Claude-Sonnet-4.5. We validate the generality of the interference across constraint types and scales. Furthermore, we identify common failure patterns, and by investigating the mechanisms of interference, we observe that failed cases allocate significantly more attention to constraints compared to successful ones. Finally, we use SUSTAINSCORE to conduct an initial investigation into how distinct post-training paradigms affect the interference, presenting empirical observations on current alignment strategies. We will release our code and data to facilitate further research

---

### PLANING: A Loosely Coupled Triangle-Gaussian Framework for Streaming 3D Reconstruction

Authors: Changjian Jiang, Kerui Ren, Xudong Li, Kaiwen Song, Guanghao Li, Linning Xu, Tao Lu, Junting Dong, Yu Zhang, Bo Dai, Mulin Yu
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.22046v4

Streaming reconstruction from monocular image sequences remains challenging, as existing methods typically favor either high-quality rendering or accurate geometry, but rarely both. We present PLANING, an efficient on-the-fly reconstruction framework built on a hybrid representation that loosely couples explicit geometric primitives with neural Gaussians, enabling geometry and appearance to be modeled in a decoupled manner. This decoupling supports an online initialization and optimization strategy that separates geometry and appearance updates, yielding stable streaming reconstruction with substantially reduced structural redundancy. PLANING improves dense mesh Chamfer-L2 by 18.52% over PGSR, surpasses ARTDECO by 1.31 dB PSNR, and reconstructs ScanNetV2 scenes in under 100 seconds, over 5x faster than 2D Gaussian Splatting, while matching the quality of offline per-scene optimization. Beyond reconstruction quality, the structural clarity and computational efficiency of PLANING make it well suited for a broad range of downstream applications, such as enabling large-scale scene modeling and simulation-ready environments for embodied AI. Project page: https://city-super.github.io/PLANING/ .

---

### Urban Neural Surface Reconstruction from Constrained Sparse Aerial Imagery with 3D SAR Fusion

Authors: Da Li, Chen Yao, Tong Mao, Jiacheng Bao, Houjun Sun
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.22045v1

Neural surface reconstruction (NSR) has recently shown strong potential for urban 3D reconstruction from multi-view aerial imagery. However, existing NSR methods often suffer from geometric ambiguity and instability, particularly under sparse-view conditions. This issue is critical in large-scale urban remote sensing, where aerial image acquisition is limited by flight paths, terrain, and cost. To address this challenge, we present the first urban NSR framework that fuses 3D synthetic aperture radar (SAR) point clouds with aerial imagery for high-fidelity reconstruction under constrained, sparse-view settings. 3D SAR can efficiently capture large-scale geometry even from a single side-looking flight path, providing robust priors that complement photometric cues from images. Our framework integrates radar-derived spatial constraints into an SDF-based NSR backbone, guiding structure-aware ray selection and adaptive sampling for stable and efficient optimization. We also construct the first benchmark dataset with co-registered 3D SAR point clouds and aerial imagery, facilitating systematic evaluation of cross-modal 3D reconstruction. Extensive experiments show that incorporating 3D SAR markedly enhances reconstruction accuracy, completeness, and robustness compared with single-modality baselines under highly sparse and oblique-view conditions, highlighting a viable route toward scalable high-fidelity urban reconstruction with advanced airborne and spaceborne optical-SAR sensing.

---

### SIA: Symbolic Interpretability for Anticipatory Deep Reinforcement Learning in Network Control

Authors: MohammadErfan Jabbari, Abhishek Duttagupta, Claudio Fiandrino, Leonardo Bonati, Salvatore D'Oro, Michele Polese, Marco Fiore, Tommaso Melodia
Published: 2026-01-29
Categories: cs.NI, cs.AI
arXiv: https://arxiv.org/abs/2601.22044v1

Deep reinforcement learning (DRL) promises adaptive control for future mobile networks but conventional agents remain reactive: they act on past and current measurements and cannot leverage short-term forecasts of exogenous KPIs such as bandwidth. Augmenting agents with predictions can overcome this temporal myopia, yet uptake in networking is scarce because forecast-aware agents act as closed-boxes; operators cannot tell whether predictions guide decisions or justify the added complexity. We propose SIA, the first interpreter that exposes in real time how forecast-augmented DRL agents operate. SIA fuses Symbolic AI abstractions with per-KPI Knowledge Graphs to produce explanations, and includes a new Influence Score metric. SIA achieves sub-millisecond speed, over 200x faster than existing XAI methods. We evaluate SIA on three diverse networking use cases, uncovering hidden issues, including temporal misalignment in forecast integration and reward-design biases that trigger counter-productive policies. These insights enable targeted fixes: a redesigned agent achieves a 9% higher average bitrate in video streaming, and SIA's online Action-Refinement module improves RAN-slicing reward by 25% without retraining. By making anticipatory DRL transparent and tunable, SIA lowers the barrier to proactive control in next-generation mobile networks.

---

### Learning to Communicate Across Modalities: Perceptual Heterogeneity in Multi-Agent Systems

Authors: Naomi Pitzer, Daniela Mihai
Published: 2026-01-29
Categories: cs.MA, cs.AI, cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2601.22041v1

Emergent communication offers insight into how agents develop shared structured representations, yet most research assumes homogeneous modalities or aligned representational spaces, overlooking the perceptual heterogeneity of real-world settings. We study a heterogeneous multi-step binary communication game where agents differ in modality and lack perceptual grounding. Despite perceptual misalignment, multimodal systems converge to class-consistent messages grounded in perceptual input. Unimodal systems communicate more efficiently, using fewer bits and achieving lower classification entropy, while multimodal agents require greater information exchange and exhibit higher uncertainty. Bit perturbation experiments provide strong evidence that meaning is encoded in a distributional rather than compositional manner, as each bit's contribution depends on its surrounding pattern. Finally, interoperability analyses show that systems trained in different perceptual worlds fail to directly communicate, but limited fine-tuning enables successful cross-system communication. This work positions emergent communication as a framework for studying how agents adapt and transfer representations across heterogeneous modalities, opening new directions for both theory and experimentation.

---

### Leviathan: Decoupling Input and Output Representations in Language Models

Authors: Reza T. Batley, Sourav Saha
Published: 2026-01-29
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.22040v2

Modern language models use a single matrix for input embedding and output projection. This couples two distinct objectives: token representation and discrimination over a vocabulary. This work introduces Leviathan, a Transformer architecture that replaces the input embedding matrix with learned embedding vectorization (LEV), a compact continuous mapping from token indices to embeddings. Leviathan's output head remains untied for a parameter increase of as low as 0.2%. Under controlled comparisons with identical Transformer backbones, Leviathan consistently improves language modeling performance over standard tied-embedding baselines across a 200M-1.2B parameter regime on The Pile with gains that grow during training. At 1.2B scale, Leviathan reduces validation perplexity by 9%, requires $2.1\times$ fewer training tokens to reach the tied baseline's final loss, and improves on all six downstream benchmarks evaluated, including a 30% reduction in LAMBADA perplexity. Frequency-stratified analysis reveals gains to be concentrated in rare tokens, where continuous parameterization reduces perplexity by 81%, falling to near zero for the most frequent.

---

### Understanding Multimodal Complementarity for Single-Frame Action Anticipation

Authors: Manuel Benavent-Lledo, Konstantinos Bacharidis, Konstantinos Papoutsakis, Antonis Argyros, Jose Garcia-Rodriguez
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.22039v1

Human action anticipation is commonly treated as a video understanding problem, implicitly assuming that dense temporal information is required to reason about future actions. In this work, we challenge this assumption by investigating what can be achieved when action anticipation is constrained to a single visual observation. We ask a fundamental question: how much information about the future is already encoded in a single frame, and how can it be effectively exploited? Building on our prior work on Action Anticipation at a Glimpse (AAG), we conduct a systematic investigation of single-frame action anticipation enriched with complementary sources of information. We analyze the contribution of RGB appearance, depth-based geometric cues, and semantic representations of past actions, and investigate how different multimodal fusion strategies, keyframe selection policies and past-action history sources influence anticipation performance. Guided by these findings, we consolidate the most effective design choices into AAG+, a refined single-frame anticipation framework. Despite operating on a single frame, AAG+ consistently improves upon the original AAG and achieves performance comparable to, or exceeding, that of state-of-the-art video-based methods on challenging anticipation benchmarks including IKEA-ASM, Meccano and Assembly101. Our results offer new insights into the limits and potential of single-frame action anticipation, and clarify when dense temporal modeling is necessary and when a carefully selected glimpse is sufficient.

---

### Optimizing Agentic Workflows using Meta-tools

Authors: Sami Abuzakuk, Anne-Marie Kermarrec, Rishi Sharma, Rasmus Moorits Veski, Martijn de Vos
Published: 2026-01-29
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.22037v2

Agentic AI enables LLM to dynamically reason, plan, and interact with tools to solve complex tasks. However, agentic workflows often require many iterative reasoning steps and tool invocations, leading to significant operational expense, end-to-end latency and failures due to hallucinations. This work introduces Agent Workflow Optimization (AWO), a framework that identifies and optimizes redundant tool execution patterns to improve the efficiency and robustness of agentic workflows. AWO analyzes existing workflow traces to discover recurring sequences of tool calls and transforms them into meta-tools, which are deterministic, composite tools that bundle multiple agent actions into a single invocation. Meta-tools bypass unnecessary intermediate LLM reasoning steps and reduce operational cost while also shortening execution paths, leading to fewer failures. Experiments on two agentic AI benchmarks show that AWO reduces the number of LLM calls up to 11.9% while also increasing the task success rate by up to 4.2 percent points.

---

### Cross-Fusion Distance: A Novel Metric for Measuring Fusion and Separability Between Data Groups in Representation Space

Authors: Xiaolong Zhang, Jianwei Zhang, Xubo Song
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22036v1

Quantifying degrees of fusion and separability between data groups in representation space is a fundamental problem in representation learning, particularly under domain shift. A meaningful metric should capture fusion-altering factors like geometric displacement between representation groups, whose variations change the extent of fusion, while remaining invariant to fusion-preserving factors such as global scaling and sampling-induced layout changes, whose variations do not. Existing distributional distance metrics conflate these factors, leading to measures that are not informative of the true extent of fusion between data groups. We introduce Cross-Fusion Distance (CFD), a principled measure that isolates fusion-altering geometry while remaining robust to fusion-preserving variations, with linear computational complexity. We characterize the invariance and sensitivity properties of CFD theoretically and validate them in controlled synthetic experiments. For practical utility on real-world datasets with domain shift, CFD aligns more closely with downstream generalization degradation than commonly used alternatives. Overall, CFD provides a theoretically grounded and interpretable distance measure for representation learning.

---

### Thinking Out of Order: When Output Order Stops Reflecting Reasoning Order in Diffusion Language Models

Authors: Longxuan Yu, Yu Fu, Shaorong Zhang, Hui Liu, Mukund Varma T, Greg Ver Steeg, Yue Dong
Published: 2026-01-29
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.22035v1

Autoregressive (AR) language models enforce a fixed left-to-right generation order, creating a fundamental limitation when the required output structure conflicts with natural reasoning (e.g., producing answers before explanations due to presentation or schema constraints). In such cases, AR models must commit to answers before generating intermediate reasoning, and this rigid constraint forces premature commitment. Masked diffusion language models (MDLMs), which iteratively refine all tokens in parallel, offer a way to decouple computation order from output structure. We validate this capability on GSM8K, Math500, and ReasonOrderQA, a benchmark we introduce with controlled difficulty and order-level evaluation. When prompts request answers before reasoning, AR models exhibit large accuracy gaps compared to standard chain-of-thought ordering (up to 67% relative drop), while MDLMs remain stable ($\leq$14% relative drop), a property we term "order robustness". Using ReasonOrderQA, we present evidence that MDLMs achieve order robustness by stabilizing simpler tokens (e.g., reasoning steps) earlier in the diffusion process than complex ones (e.g., final answers), enabling reasoning tokens to stabilize before answer commitment. Finally, we identify failure conditions where this advantage weakens, outlining the limits required for order robustness.

---

### Holographic generative flows with AdS/CFT

Authors: Ehsan Mirafzali, Sanjit Shashi, Sanya Murdeshwar, Edgar Shaghoulian, Daniele Venturi, Razvan Marinescu
Published: 2026-01-29
Categories: cs.LG, gr-qc, hep-th
arXiv: https://arxiv.org/abs/2601.22033v2

Holography, in the form of the anti-de Sitter/conformal field theory (AdS/CFT) correspondence, offers a natural setting for generative modelling. Data on a boundary manifold lifts into a higher-dimensional bulk through a propagator, and this extra dimension plays the role of a flow parameter. We exploit this structure to build GenAdS, an approach to generative flow matching in which the dynamics are represented by the evolution of fields in AdS, together with a residual correction learned by a neural network. Boundary samples are encoded as scalar sources, transported into the bulk along the flow, and decoded after numerical integration. Our paradigm combines a Fourier-space encoding scheme for the data as AdS sources, a normalised radial phase space in which to stage the flow-matching dynamics, and a Klein--Gordon backbone to guide the flow. On a two-dimensional checkerboard benchmark, our experiments show that most of the benefit of GenAdS comes from the Fourier representation and convolutional architecture. However, when we remove momentum-channel regularisation, our most physics-informed GenAdS variant rivals the strongest physics-free control on boundary violation. On MNIST, GenAdS models remain close to a convolutional baseline on fidelity while achieving significantly higher recall at comparable precision, suggesting a fidelity-coverage trade-off. Our findings establish GenAdS as a physically interpretable and experimentally controllable framework for generative modelling, with many avenues for future extension.

---

### Drive-JEPA: Video JEPA Meets Multimodal Trajectory Distillation for End-to-End Driving

Authors: Linhan Wang, Zichong Yang, Chen Bai, Guoxiang Zhang, Xiaotong Liu, Xiaoyin Zheng, Xiao-Xiao Long, Chang-Tien Lu, Cheng Lu
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.22032v2

End-to-end autonomous driving increasingly leverages self-supervised video pretraining to learn transferable planning representations. However, pretraining video world models for scene understanding has so far brought only limited improvements. This limitation is compounded by the inherent ambiguity of driving: each scene typically provides only a single human trajectory, making it difficult to learn multimodal behaviors. In this work, we propose Drive-JEPA, a framework that integrates Video Joint-Embedding Predictive Architecture (V-JEPA) with multimodal trajectory distillation for end-to-end driving. First, we adapt V-JEPA for end-to-end driving, pretraining a ViT encoder on large-scale driving videos to produce predictive representations aligned with trajectory planning. Second, we introduce a proposal-centric planner that distills diverse simulator-generated trajectories alongside human trajectories, with a momentum-aware selection mechanism to promote stable and safe behavior. When evaluated on NAVSIM, the V-JEPA representation combined with a simple transformer-based decoder outperforms prior methods by 3 PDMS in the perception-free setting. The complete Drive-JEPA framework achieves 93.3 PDMS on v1 and 87.8 EPDMS on v2, setting a new state-of-the-art.

---

### Causal Autoregressive Diffusion Language Model

Authors: Junhao Ruan, Bei Li, Yongjing Yin, Pengcheng Huang, Xin Chen, Jingang Wang, Xunliang Cai, Tong Xiao, JingBo Zhu
Published: 2026-01-29
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.22031v1

In this work, we propose Causal Autoregressive Diffusion (CARD), a novel framework that unifies the training efficiency of ARMs with the high-throughput inference of diffusion models. CARD reformulates the diffusion process within a strictly causal attention mask, enabling dense, per-token supervision in a single forward pass. To address the optimization instability of causal diffusion, we introduce a soft-tailed masking schema to preserve local context and a context-aware reweighting mechanism derived from signal-to-noise principles. This design enables dynamic parallel decoding, where the model leverages KV-caching to adaptively generate variable-length token sequences based on confidence. Empirically, CARD outperforms existing discrete diffusion baselines while reducing training latency by 3 $\times$ compared to block diffusion methods. Our results demonstrate that CARD achieves ARM-level data efficiency while unlocking the latency benefits of parallel generation, establishing a robust paradigm for next-generation efficient LLMs.

---

### Per-parameter Task Arithmetic for Unlearning in Large Language Models

Authors: Chengyi Cai, Zesheng Ye, Jiangchao Yao, Jianzhong Qi, Bo Han, Xiaolu Zhang, Feng Liu, Jun Zhou
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22030v1

In large language model (LLM) unlearning, private information is required to be removed. Task arithmetic unlearns by subtracting a specific task vector (TV)--defined as the parameter difference between a privacy-information-tuned model and the original model. While efficient, it can cause over-forgetting by disrupting parameters essential for retaining other information. Motivated by the observation that each parameter exhibits different importance for forgetting versus retention, we propose a per-parameter task arithmetic (PerTA) mechanism to rescale the TV, allowing per-parameter adjustment. These weights quantify the relative importance of each parameter for forgetting versus retention, estimated via gradients (i.e., PerTA-grad) or the diagonal Fisher information approximation (i.e., PerTA-fisher). Moreover, we discuss the effectiveness of PerTA, extend it to a more general form, and provide further analysis. Extensive experiments demonstrate that PerTA consistently improves upon standard TV, and in many cases surpasses widely used training-based unlearning methods in both forgetting effectiveness and overall model utility. By retaining the efficiency of task arithmetic while mitigating over-forgetting, PerTA offers a principled and practical framework for LLM unlearning.

---

### Inverse Problems Conditioned on Observation Ensembles: Applications and Methods

Authors: Zhengyan Huan, Camila Pazos, Martin Klassen, Vincent Croft, Pierre-Hugues Beauchemin, Shuchin Aeron
Published: 2026-01-29
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2601.22029v2

We introduce a new multivariate statistical problem that we refer to as the Ensemble-conditioned Inverse Problem (EIP). The aim of EIP is to invert for an ensemble that is distributed according to the pushforward of a prior under a forward process. In high energy physics (HEP), this is related to a widely known problem called unfolding, which aims to reconstruct the true physics distribution from observations that are distorted by detector effects. The EIP also arises in full waveform inversion (FWI) and inverse imaging with unknown priors. We propose non-iterative inference-time methods that construct posterior samplers based on a new class of conditional generative models, which we call ensemble inverse generative models. For the posterior modeling, these models additionally use the ensemble information contained in the observation set on top of single observations. Unlike existing methods, our proposed methods avoid explicit and iterative use of the forward model at inference time via training across several sets of truth-observation pairs that are consistent with the same forward model, but originate from a wide range of priors. We empirically demonstrate that this training procedure can implicitly encode the likelihood model, enabling direct posterior inference for unseen priors to some degree. We benchmark the proposed method on several synthetic and real datasets in inverse imaging, HEP, and FWI. Our code is available at https://github.com/ZhengyanHuan/EIP.

---

### From Logits to Latents: Contrastive Representation Shaping for LLM Unlearning

Authors: Haoran Tang, Rajiv Khanna
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22028v1

Most LLM unlearning methods aim to approximate retrain-from-scratch behaviors with minimal distribution shift, often via alignment-style objectives defined in the prediction space. While effective at reducing forgotten content generation, such approaches may act as suppression: forgotten concepts can persist in representations and remain entangled with retained knowledge. We introduce CLReg, a contrastive representation regularizer that identifies forget features while pushing them away from retain features, explicitly reducing forget-retain interference with minimal shifts on retain features. We provide first theoretical insights that relate representation shaping to entanglement reduction. Across unlearning benchmarks and LLMs of different sizes, CLReg decreases forget-retain representation entanglement that facilitates mainstream unlearning methods without positing extra privacy risks, inspiring future work that reshapes the representation space to remove forget concepts.

---

### CAR-bench: Evaluating the Consistency and Limit-Awareness of LLM Agents under Real-World Uncertainty

Authors: Johannes Kirmayr, Lukas Stappen, Elisabeth André
Published: 2026-01-29
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.22027v1

Existing benchmarks for Large Language Model (LLM) agents focus on task completion under idealistic settings but overlook reliability in real-world, user-facing applications. In domains, such as in-car voice assistants, users often issue incomplete or ambiguous requests, creating intrinsic uncertainty that agents must manage through dialogue, tool use, and policy adherence. We introduce CAR-bench, a benchmark for evaluating consistency, uncertainty handling, and capability awareness in multi-turn, tool-using LLM agents in an in-car assistant domain. The environment features an LLM-simulated user, domain policies, and 58 interconnected tools spanning navigation, productivity, charging, and vehicle control. Beyond standard task completion, CAR-bench introduces Hallucination tasks that test agents' limit-awareness under missing tools or information, and Disambiguation tasks that require resolving uncertainty through clarification or internal information gathering. Baseline results reveal large gaps between occasional and consistent success on all task types. Even frontier reasoning LLMs achieve less than 50% consistent pass rate on Disambiguation tasks due to premature actions, and frequently violate policies or fabricate information to satisfy user requests in Hallucination tasks, underscoring the need for more reliable and self-aware LLM agents in real-world settings.

---

### Hybrid Foveated Path Tracing with Peripheral Gaussians for Immersive Anatomy

Authors: Constantin Kleinbeck, Luisa Theelke, Hannah Schieber, Ulrich Eck, Rüdiger von Eisenhart-Rothe, Daniel Roth
Published: 2026-01-29
Categories: cs.GR, cs.CV
arXiv: https://arxiv.org/abs/2601.22026v1

Volumetric medical imaging offers great potential for understanding complex pathologies. Yet, traditional 2D slices provide little support for interpreting spatial relationships, forcing users to mentally reconstruct anatomy into three dimensions. Direct volumetric path tracing and VR rendering can improve perception but are computationally expensive, while precomputed representations, like Gaussian Splatting, require planning ahead. Both approaches limit interactive use. We propose a hybrid rendering approach for high-quality, interactive, and immersive anatomical visualization. Our method combines streamed foveated path tracing with a lightweight Gaussian Splatting approximation of the periphery. The peripheral model generation is optimized with volume data and continuously refined using foveal renderings, enabling interactive updates. Depth-guided reprojection further improves robustness to latency and allows users to balance fidelity with refresh rate. We compare our method against direct path tracing and Gaussian Splatting. Our results highlight how their combination can preserve strengths in visual quality while re-generating the peripheral model in under a second, eliminating extensive preprocessing and approximations. This opens new options for interactive medical visualization.

---

### When Generic Prompt Improvements Hurt: Evaluation-Driven Iteration for LLM Applications

Authors: Daniel Commey
Published: 2026-01-29
Categories: cs.CL, cs.AI, cs.IR, cs.SE
arXiv: https://arxiv.org/abs/2601.22025v2

Evaluating Large Language Model (LLM) applications differs from conventional software testing because outputs are probabilistic, semantically variable, and sensitive to prompt and model changes. This technical report proposes the Minimum Viable Evaluation Suite (MVES), an audit-oriented structure for application-level LLM evaluation. MVES links application categories to failure modes, metrics, required artifacts, and validation evidence across general LLM applications, retrieval-augmented systems, and agentic workflows. We pair the framework with a reproducible local evaluation harness covering structured extraction, RAG citation/content-compliance, and instruction-following checks. Using Ollama with Llama 3 8B Instruct and Qwen 2.5 7B Instruct, we evaluate five prompt conditions over expanded 30-case-per-suite ablations. The results show that, in the tested local conditions, generic prompt additions do not produce monotonic improvements: stronger output-contract prompts improve strict extraction for both models, while RAG citation/content-compliance declines under some generic-rule conditions. The largest observed decline occurs for Qwen 2.5 on RAG when generic rules are appended to the user prompt, from 26/30 to 9/30. These findings support evaluation-driven prompt iteration: prompt changes should be treated as potential regression risks and tested against task-specific suites before deployment. The accompanying repository contains the test suites, prompt variants, evaluation harness, raw result logs, and scripts needed to reproduce the reported local ablations.

---

### SymbXRL: Symbolic Explainable Deep Reinforcement Learning for Mobile Networks

Authors: Abhishek Duttagupta, MohammadErfan Jabbari, Claudio Fiandrino, Marco Fiore, Joerg Widmer
Published: 2026-01-29
Categories: cs.NI, cs.AI
arXiv: https://arxiv.org/abs/2601.22024v1

The operation of future 6th-generation (6G) mobile networks will increasingly rely on the ability of deep reinforcement learning (DRL) to optimize network decisions in real-time. DRL yields demonstrated efficacy in various resource allocation problems, such as joint decisions on user scheduling and antenna allocation or simultaneous control of computing resources and modulation. However, trained DRL agents are closed-boxes and inherently difficult to explain, which hinders their adoption in production settings. In this paper, we make a step towards removing this critical barrier by presenting SymbXRL, a novel technique for explainable reinforcement learning (XRL) that synthesizes human-interpretable explanations for DRL agents. SymbXRL leverages symbolic AI to produce explanations where key concepts and their relationships are described via intuitive symbols and rules; coupling such a representation with logical reasoning exposes the decision process of DRL agents and offers more comprehensible descriptions of their behaviors compared to existing approaches. We validate SymbXRL in practical network management use cases supported by DRL, proving that it not only improves the semantics of the explanations but also paves the way for explicit agent control: for instance, it enables intent-based programmatic action steering that improves by 12% the median cumulative reward over a pure DRL solution.

---

### Visual-Guided Key-Token Regularization for Multimodal Large Language Model Unlearning

Authors: Chengyi Cai, Zesheng Ye, Peike Li, Bo Han, Jianzhong Qi, Feng Liu
Published: 2026-01-29
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2601.22020v1

Unlearning in Multimodal Large Language Models (MLLMs) prevents the model from revealing private information when queried about target images. Existing MLLM unlearning methods largely adopt approaches developed for LLMs. They treat all answer tokens uniformly, disregarding their varying importance in the unlearning process. Moreover, these methods focus exclusively on the language modality, disregarding visual cues that indicate key tokens in answers. In this paper, after formulating the problem of unlearning in multimodal question answering for MLLMs, we propose Visual-Guided Key-Token Regularization (ViKeR). We leverage irrelevant visual inputs to predict ideal post-unlearning token-level distributions and use these distributions to regularize the unlearning process, thereby prioritizing key tokens. Further, we define key tokens in unlearning via information entropy and discuss ViKeR's effectiveness through token-level gradient reweighting, which amplifies updates on key tokens. Experiments on MLLMU and CLEAR benchmarks demonstrate that our method effectively performs unlearning while mitigating forgetting and maintaining response coherence.

---

### TBDFiltering: Sample-Efficient Tree-Based Data Filtering

Authors: Robert Istvan Busa-Fekete, Julian Zimmert, Anne Xiangyi Zheng, Claudio Gentile, Andras Gyorgy
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22016v1

The quality of machine learning models depends heavily on their training data. Selecting high-quality, diverse training sets for large language models (LLMs) is a difficult task, due to the lack of cheap and reliable quality metrics. While querying existing LLMs for document quality is common, this is not scalable to the large number (billions) of documents used in training. Instead, practitioners often use classifiers trained on sparse quality signals. In this paper, we propose a text-embedding-based hierarchical clustering approach that adaptively selects the documents to be evaluated by the LLM to estimate cluster quality. We prove that our method is query efficient: under the assumption that the hierarchical clustering contains a subtree such that each leaf cluster in the tree is pure enough (i.e., it mostly contains either only good or only bad documents), with high probability, the method can correctly predict the quality of each document after querying a small number of documents. The number of such documents is proportional to the size of the smallest subtree with (almost) pure leaves, without the algorithm knowing this subtree in advance. Furthermore, in a comprehensive experimental study, we demonstrate the benefits of our algorithm compared to other classifier-based filtering methods.

---

### Vidmento: Creating Video Stories Through Context-Aware Expansion With Generative Video

Authors: Catherine Yeh, Anh Truong, Mira Dontcheva, Bryan Wang
Published: 2026-01-29
Categories: cs.HC, cs.AI, cs.MM
arXiv: https://arxiv.org/abs/2601.22013v2

Video storytelling is often constrained by available material, limiting creative expression and leaving undesired narrative gaps. Generative video offers a new way to address these limitations by augmenting captured media with tailored visuals. To explore this potential, we interviewed eight video creators to identify opportunities and challenges in integrating generative video into their workflows. Building on these insights and established filmmaking principles, we developed Vidmento, a tool for authoring hybrid video stories that combine captured and generated media through context-aware expansion. Vidmento surfaces opportunities for story development, generates clips that blend stylistically and narratively with surrounding media, and provides controls for refinement. In a study with 12 creators, Vidmento supported narrative development and exploration by systematically expanding initial materials with generative media, enabling expressive video storytelling aligned with creative intent. We highlight how creators bridge story gaps with generative content and where they find this blending capability most valuable.

---

### Putting a Face to Forgetting: Continual Learning meets Mechanistic Interpretability

Authors: Sergi Masip, Gido M. van de Ven, Javier Ferrando, Tinne Tuytelaars
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22012v3

Catastrophic forgetting in continual learning is often measured at the performance or last-layer representation level, overlooking the underlying mechanisms. We introduce a mechanistic framework that offers a geometric interpretation of catastrophic forgetting as the result of transformations to the encoding of individual features. These transformations can lead to forgetting by reducing the allocated capacity of features or by disrupting their readout by downstream computations. Analysis of a tractable toy model formalizes this view, allowing us to identify best- and worst-case scenarios. Through experiments on this model, we empirically test our formal analysis and highlight the detrimental effect of depth. Finally, we demonstrate how our framework can be used in the analysis of practical models through the use of Crosscoders. We do so through a case study example of a Vision Transformer trained on sequential CIFAR-10. Our work provides a new, feature-centric vocabulary for continual learning.

---

### Exploring Diverse Generation Paths via Inference-time Stiefel Activation Steering

Authors: Dongxuan Zhu, Ly Tran Ho Khanh, Andy Yat-Ming Cheung, Man-Chung Yue, Viet Anh Nguyen
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.22010v1

Language models often default to a narrow set of high-probability outputs, leaving their generation paths homogeneous and prone to mode collapse. Sampling-based strategies inject randomness but still struggle to guarantee diversity across multiple concurrent generation runs. We address this limitation by introducing STARS ($\textbf{St}$iefel-based $\textbf{A}$ctivation Steering for Diverse $\textbf{R}$ea$\textbf{S}$oning), a training-free, inference-time intervention method that transforms activation steering into an exploration engine. At each token, STARS collects the hidden activations of concurrent generation runs and optimizes multiple additive steering directions jointly on the Stiefel manifold. STARS maximizes the geometric volume of the steered activations, while the Stiefel manifold induces orthogonality of the steering interventions. This formulation explicitly promotes divergent activation vectors of concurrent generation runs, and implicitly promotes divergent generation trajectories. This manifold optimization formulation can be solved using a Riemannian gradient descent algorithm with convergence guarantees, but this algorithm is too time-consuming for real-time inference. To guarantee low latency, we further design a lightweight one-step update with an aggressive, closed-form stepsize. For test case generation and scientific discovery benchmarks, STARS consistently outperforms standard sampling methods, achieving greater diversity without sacrificing qualitative performance.

---

### MEIDNet: Multimodal generative AI framework for inverse materials design

Authors: Anand Babu, Rogério Almeida Gouvêa, Pierre Vandergheynst, Gian-Marco Rignanese
Published: 2026-01-29
Categories: cond-mat.mtrl-sci, cs.AI, cs.LG, physics.comp-ph
arXiv: https://arxiv.org/abs/2601.22009v1

In this work, we present Multimodal Equivariant Inverse Design Network (MEIDNet), a framework that jointly learns structural information and materials properties through contrastive learning, while encoding structures via an equivariant graph neural network (EGNN). By combining generative inverse design with multimodal learning, our approach accelerates the exploration of chemical-structural space and facilitates the discovery of materials that satisfy predefined property targets. MEIDNet exhibits strong latent-space alignment with cosine similarity 0.96 by fusion of three modalities through cross-modal learning. Through implementation of curriculum learning strategies, MEIDNet achieves ~60 times higher learning efficiency than conventional training techniques. The potential of our multimodal approach is demonstrated by generating low-bandgap perovskite structures at a stable, unique, and novel (SUN) rate of 13.6 %, which are further validated by ab initio methods. Our inverse design framework demonstrates both scalability and adaptability, paving the way for the universal learning of chemical space across diverse modalities.

---

### Hierarchy of discriminative power and complexity in learning quantum ensembles

Authors: Jian Yao, Pengtao Li, Xiaohui Chen, Quntao Zhuang
Published: 2026-01-29
Categories: quant-ph, math.ST, stat.ML
arXiv: https://arxiv.org/abs/2601.22005v1

Distance metrics are central to machine learning, yet distances between ensembles of quantum states remain poorly understood due to fundamental quantum measurement constraints. We introduce a hierarchy of integral probability metrics, termed MMD-$k$, which generalizes the maximum mean discrepancy to quantum ensembles and exhibit a strict trade-off between discriminative power and statistical efficiency as the moment order $k$ increases. For pure-state ensembles of size $N$, estimating MMD-$k$ using experimentally feasible SWAP-test-based estimators requires $Θ(N^{2-2/k})$ samples for constant $k$, and $Θ(N^3)$ samples to achieve full discriminative power at $k = N$. In contrast, the quantum Wasserstein distance attains full discriminative power with $Θ(N^2 \log N)$ samples. These results provide principled guidance for the design of loss functions in quantum machine learning, which we illustrate in the training quantum denoising diffusion probabilistic models.

---

### Efficient Stochastic Optimisation via Sequential Monte Carlo

Authors: James Cuin, Davide Carbone, Yanbo Tang, O. Deniz Akyildiz
Published: 2026-01-29
Categories: stat.ML, cs.LG, stat.CO
arXiv: https://arxiv.org/abs/2601.22003v2

The problem of optimising functions with intractable gradients frequently arises in machine learning and statistics, ranging from maximum marginal likelihood estimation procedures to fine-tuning of generative models. Stochastic approximation methods for this class of problems typically require inner sampling loops to obtain (biased) stochastic gradient estimates, which rapidly becomes computationally expensive. In this work, we develop sequential Monte Carlo (SMC) samplers for optimisation of functions with intractable gradients. Our approach replaces expensive inner sampling methods with efficient SMC approximations, which can result in significant computational gains. We establish convergence results for the basic recursions defined by our methodology which SMC samplers approximate. We demonstrate the effectiveness of our approach on the reward-tuning of energy-based models within various settings.

---

### Understanding Rate-Distortion Performance in Distributed Transformer Inference

Authors: Anderson de Andrade, Alon Harell, Ivan V. Bajić
Published: 2026-01-29
Categories: cs.LG, cs.IT
arXiv: https://arxiv.org/abs/2601.22002v5

Transformers achieve superior performance on many tasks, but impose heavy compute and memory requirements during inference. This inference can be made more efficient by partitioning the process across multiple devices, which, in turn, requires compressing its intermediate representations. We study compressibility of transformer's intermediate representations via learned compression through the lens of rate-distortion and the theory of usable information. Our study reveals that, unlike convolutional models, deeper representations in transformers become more difficult to compress. The reason for this behavior is twofold: first, the complexity of representation increases as we move deeper into the transformer, necessitating a higher rate; and second, the higher complexity of representations worsens the generaliztion bound for learned entropy estimates, which further compromises compression performance. Through a combination of experiments and theory, we characterize and analyze the compressibility of transformer representations, derive bounds on the achievable rate of learned codecs applied to these representations, and offer a unified lens for understanding rate-distortion performance in representation coding.

---

### Heterogeneous Computing: The Key to Powering the Future of AI Agent Inference

Authors: Yiren Zhao, Junyi Liu
Published: 2026-01-29
Categories: cs.AI, cs.AR, cs.DC
arXiv: https://arxiv.org/abs/2601.22001v1

AI agent inference is driving an inference heavy datacenter future and exposes bottlenecks beyond compute - especially memory capacity, memory bandwidth and high-speed interconnect. We introduce two metrics - Operational Intensity (OI) and Capacity Footprint (CF) - that jointly explain regimes the classic roofline analysis misses, including the memory capacity wall. Across agentic workflows (chat, coding, web use, computer use) and base model choices (GQA/MLA, MoE, quantization), OI/CF can shift dramatically, with long context KV cache making decode highly memory bound. These observations motivate disaggregated serving and system level heterogeneity: specialized prefill and decode accelerators, broader scale up networking, and decoupled compute-memory enabled by optical I/O. We further hypothesize agent-hardware co design, multiple inference accelerators within one system, and high bandwidth, large capacity memory disaggregation as foundations for adaptation to evolving OI/CF. Together, these directions chart a path to sustain efficiency and capability for large scale agentic AI inference.

---

### Negatives-Dominant Contrastive Learning for Generalization in Imbalanced Domains

Authors: Meng Cao, Jiexi Liu, Songcan Chen
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.21999v1

Imbalanced Domain Generalization (IDG) focuses on mitigating both domain and label shifts, both of which fundamentally shape the model's decision boundaries, particularly under heterogeneous long-tailed distributions across domains. Despite its practical significance, it remains underexplored, primarily due to the technical complexity of handling their entanglement and the paucity of theoretical foundations. In this paper, we begin by theoretically establishing the generalization bound for IDG, highlighting the role of posterior discrepancy and decision margin. This bound motivates us to focus on directly steering decision boundaries, marking a clear departure from existing methods. Subsequently, we technically propose a novel Negative-Dominant Contrastive Learning (NDCL) for IDG to enhance discriminability while enforce posterior consistency across domains. Specifically, inter-class decision-boundary separation is enhanced by placing greater emphasis on negatives as the primary signal in our contrastive learning, naturally amplifying gradient signals for minority classes to avoid the decision boundary being biased toward majority classes. Meanwhile, intra-class compactness is encouraged through a re-weighted cross-entropy strategy, and posterior consistency across domains is enforced through a prediction-central alignment strategy. Finally, rigorous yet challenging experiments on benchmarks validate the effectiveness of our NDCL. The code is available at https://github.com/Alrash/NDCL.

---

### Causal World Modeling for Robot Control

Authors: Lin Li, Qihang Zhang, Yiming Luo, Shuai Yang, Ruilin Wang, Fei Han, Mingrui Yu, Zelin Gao, Nan Xue, Xing Zhu, Yujun Shen, Yinghao Xu
Published: 2026-01-29
Categories: cs.CV, cs.RO
arXiv: https://arxiv.org/abs/2601.21998v2

This work highlights that video world modeling, alongside vision-language pre-training, establishes a fresh and independent foundation for robot learning. Intuitively, video world models provide the ability to imagine the near future by understanding the causality between actions and visual dynamics. Inspired by this, we introduce LingBot-VA, an autoregressive diffusion framework that learns frame prediction and policy execution simultaneously. Our model features three carefully crafted designs: (1) a shared latent space, integrating vision and action tokens, driven by a Mixture-of-Transformers (MoT) architecture, (2) a closed-loop rollout mechanism, allowing for ongoing acquisition of environmental feedback with ground-truth observations, (3) an asynchronous inference pipeline, parallelizing action prediction and motor execution to support efficient control. We evaluate our model on both simulation benchmarks and real-world scenarios, where it shows significant promise in long-horizon manipulation, data efficiency in post-training, and strong generalizability to novel configurations. The code and model are made publicly available to facilitate the community.

---