AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Toward Optimal Adenovirus Detection Using YOLO26

Olivier Rukundo

This study systematically benchmarks different data augmentation setups across the baseline YOLO26 model size variants to determine the most effective setup for adenovirus detection in TEM images. The benchmarked setups include NAS, GAS, GMAS, and DAS, all evaluated under identical training conditions. A modified YOLO26 model leveraging P2, expanded STAL, increased topk, and MuSGD was also tested across the same benchmarked setups. The adenovirus dataset, selected from the published TEM virus dataset, was re-annotated by leveraging adenovirus particle positions to generate YOLO-compatible bounding box annotations. The modifications produced their largest performance gains under GAS and GMAS, with modified YOLO26x trained using GAS achieving a mAP@50 of 0.80, Precision of 0.81 and a Recall of 0.80

arXiv

Exploratory, Communicative, and Deployable: Vision-Driven Embodied Agents for Open-World Mobile Manipulation

Boyu Mi, Mengchen Ma, Yifei Yao, Xing Gao, Junting Chen, Yangzi Li, Zihou Zhu, Guohao Li, Zhenfei Yin, Tai Wang, Yao Mu, Jiangmiao Pang, Hanqing Wang

Real-world deployment of embodied agents requires active exploration, visual grounding, and interactive intent disambiguation. However, existing frameworks often rely on privileged simulator states or assume complete instructions, bypassing realistic deployment challenges. To bridge this gap, we present REAL, an agentic framework for open-world mobile manipulation. REAL establishes sim-to-real-consistent environment APIs without oracle perception and integrates a simulated user to enable human-in-the-loop interaction. Within this environment, we design diverse task compositions to drive data collection, supervised fine-tuning, and online reinforcement learning, systematically optimizing agent performance. To comprehensively evaluate this approach, we introduce REAL-Bench, a benchmark spanning 241 tasks across active exploration, visual distraction, articulated manipulation, and interactive disambiguation. Experimental results demonstrate that our trained agent outperforms leading commercial closed-source VLMs on interactive tasks with a 56.9% success rate. Further empirical analysis reveals that our hierarchical training pipeline successfully aligns the model's tool-use capabilities while maintaining robust open-vocabulary reasoning under extended exploration horizons. Finally, we deploy and evaluate our framework on a physical dual-arm mobile robot, where it achieves a 78.3% end-to-end success rate over 60 real-world episodes. These physical trials demonstrate robust zero-shot transferability to unseen household scenarios, validating that our sim-to-real-consistent design successfully bridges the reality gap for long-horizon mobile manipulation. Code is available at https://github.com/InternRobotics/REAL.

arXiv

IB-Flow: Information Bottleneck-Guided CFG Distillation for Few-Step Text-to-Image Generation

Yiting Wang, Jingyi Zhang, Wenhu Zhang, Ke Chao, Yves Liang, Kun Cheng, Kang Zhao

While large-scale text-to-image generative models have achieved unprecedented visual performance, their inherent reliance on multi-step iterative solvers incurs severe inference latency. Few-step distillation targeting the Classifier-Free Guidance (CFG) trajectory has emerged as the prevalent dual-dimensional compression paradigm. However, existing frameworks remain subjugated by a coarse-grained blind injection paradigm that perpetually enforces a globally static guidance strength while indiscriminately sampling the supervisor timestep. This state-agnostic design completely disregards the intrinsic nature of image generation as a dynamic evolutionary process characterized by progressive entropy reduction, which not only restricts the performance boundary of few-step compression but also precipitates severe CFG over-conditioning artifacts. To transcend these limitations, we re-examine the distillation procedure through the theoretical lens of Information Theory, formally modeling it as a dynamic mutual information game constrained by the Information Bottleneck (IB) principle. Specifically, we dismantle traditional blind assumptions via a dual-track adaptive framework. To determine the injection target, we propose an instance-aware selection mechanism that transmutes the intractable KL divergence constraint into a zero-overhead closed-form solution predicated on the local vector field norm. To regulate the injection strength, we introduce an entropy-aware schedule that dynamically decays alongside the SNR, applying maximal thrust for initial structural anchoring before smoothly reverting to the natural manifold to refine micro-details. Extensive empirical evaluations corroborate that our framework fundamentally eradicates over-conditioning artifacts, shattering the performance ceiling to achieve SOTA generative fidelity under extremely stringent 2-step configurations.

arXiv

SLAM: Structured and Localized Analytic Manifold Adaptation for Forgetting-Immune and Domain-Robust Lifelong VPR

Kenta Tsukahara, Kanji Tanaka, Rai Hisada

Visual Place Recognition (VPR) under long-term operation is essential for autonomous mobile robots. While Analytic Class-Incremental Learning (ACIL) provides memory-free ($O(1)$) task adaptation with exact forgetting immunity, applying it to lifelong VPR suffers from extreme vulnerability to non-linear domain shifts induced by environmental variations. In this work, we introduce the concept of the \textbf{ACIL-Domain (ACIL-D)}---a canonical invariant feature manifold where autocorrelation states remain locked. We resolve the domain vulnerability via Disentangled Domain Alignment (D-DA), which decouples latent features into invariant semantics within ACIL-D and variant style vectors for directional projection. Furthermore, by uncovering an algebraic isomorphism between recursive ACIL updates and Extended Kalman Filter (EKF) covariance propagation, we establish a control-theoretic framework designated as \textbf{SLAM} (\textbf{S}tructured and \textbf{L}ocalized \textbf{A}nalytic \textbf{M}anifold adaptation). SLAM integrates dynamic temperature-scaled Gaussian Mixture Models (GMM) to isolate topological non-linearities, Unscented perturbed propagation to dampen feature variations, and minimax $H_{\infty}$-robust criteria to bound worst-case noise accumulation. Empirical evaluations on the non-stationary NCLT dataset demonstrate that our proposed framework substantially outperforms existing baselines, achieving a final all-class accuracy of 27.7\% with the full SLAM framework (and up to 29.0\% with the U+H variant) while guaranteeing complete forgetting immunity.

arXiv

Enhancing Pathological VLMs with Cross-scale Reasoning

Chi Phan, Tianyi Zhang, Qiaochu Xue, Yufeng Wu, Dan Hu, Zeyu Liu, Sudong Wang, Yueming Jin

Pathological images are inherently multi-scale, requiring pathologists to integrate evidence from global tissue architecture at low magnification to cellular morphology at higher magnification for accurate diagnosis. While existing pathological datasets for vision-language models (VLMs) include various scales, they often lack explicit cross-scale reasoning objectives. This limitation prevents VLMs from capturing essential cross-scale representations and learning evidence-based reasoning. To bridge this gap, we introduce the first cross-scale training and evaluation paradigm that formulates pathology interpretation as multi-magnification reasoning. However, creating such a task reveals a critical challenge: multi-image visual question answering (VQA) is prone to text-only shortcuts, which allow models to guess answers using magnification-dependent artifacts rather than visual evidence. To address this, we propose a leakage-aware curation pipeline that combines adversarial text-only screening with constraint-guided question design. Using this pipeline, we construct Scale-VQA, a high-quality benchmark with 4,685 multiple-choice questions grounded in 2,537 pathology images across multiple magnification levels. Finally, we present ScaleReasoner-R1, a model trained via reinforcement learning to optimize performance on cross-scale VQA tasks. ScaleReasoner-R1 achieves state-of-the-art performance on our cross-scale reasoning benchmark and generalizes to SOTA performance on established single-scale benchmarks. Findings suggest that even the limited cross-scale supervision can significantly improve pathological understanding. Code is available at https://github.com/iMVR-PL/ScaleReasoner-R1.

arXiv

How Much MRI Preprocessing Is Enough? A Cost-Utility Study for Brain MRI Foundation Models

Jiangshuan Pang, Wangyang Tang, Jing Yan, Zhixuan Cheng, Youzhe He, Yiting Deng, Zhenkun Zhuang, Shiping Liu

MRI preprocessing defines the input distribution seen by brain MRI foundation models, yet it is usually treated as routine data cleaning rather than a modeling choice. We ask how much preprocessing is worth its computational cost for self-supervised 3D MRI pretraining. Keeping the corpus, 3D ViT backbone, masking protocol, and downstream evaluations fixed, we compare a graded P0-P7 preprocessing spectrum for masked autoencoding (MAE) and joint-embedding predictive learning (JEPA) on 20,000 heterogeneous brain MRI volumes, then transfer the encoders to IDH prediction, MCI classification, brain age regression, and GLI/PED tumor segmentation. The results do not support a simple "more is better" rule. P0/P1 are numerically unstable, making P2 the lowest-cost feasible level; beyond P2, choosing the best feasible preprocessing level improves aggregate utility by only 3.4 percentage points for MAE and 1.8 percentage points for JEPA, with most paired gains statistically unresolved. Stronger preprocessing is beneficial only in selected regimes: IDH improves modestly, AGE and GLI/PED are often near or best at P2, and MCI shows the clearest empirical P7 gain. Cross-level MCI transfer further shows that much of the P7 advantage can be recovered by applying stronger preprocessing downstream, without requiring P7 throughout pretraining. These findings recast MRI preprocessing as a downstream-aware cost-utility decision rather than a default escalation pipeline. Code is available at https://github.com/PangJiangShuan/PreBrain.

arXiv

FLAME: Physics-Guided Neural Operators for Onboard Satellite Methane Detection in Hyperspectral Imagery

Junhyuk Heo, Junghwan Park, Sangcheol Sim, Beomkyu Choi, Woojin Cho

Methane is a major driver of near-term climate change, and rapidly identifying its emission sources is a critical climate intervention. Spaceborne hyperspectral imagery is the primary tool for this task, but the volume of data produced by each sensor makes ground-based detection impractical and necessitates onboard detection. Classical methods incur prohibitive computational cost on onboard hardware, while deep learning models are fast but fall short on detection quality. We propose FLAME, a physics-guided neural operator that builds the physics of methane absorption directly into its architecture. On the methane detection benchmark, FLAME achieves the highest detection accuracy among all evaluated methods, reduces the pixel-level false positive rate by nearly $3\times$ over the strongest neural baseline, uses the fewest parameters among learned baselines, and runs within the latency budget of onboard satellite hardware.

arXiv

Task-Aligned Self-Supervised Learning for Medical Image Analysis: A Task-Oriented Review with Practical Design Guidelines

Chathura Wimalasiri, Yuchong Yao, Kishor Nandakishor, Marimuthu Palaniswami

Self-supervised learning (SSL) is increasingly used in medical image analysis to reduce dependence on costly expert annotations by learning transferable representations from unlabeled data. However, SSL performance depends not only on model architecture but also on whether the self-supervised objective preserves the information required by the downstream clinical task. This review presents a task-oriented synthesis of SSL methods for medical imaging, focusing on how the design of the self-supervised objective interacts with imaging modality, label availability, and downstream performance. We analyze $78$ studies published from 2017 to 2025 and organize them into four paradigms: contrastive, non-contrastive and predictive, generative and reconstruction-based, and hybrid learning. Rather than cataloging methods chronologically, we examine how these paradigms support classification, segmentation, detection, reconstruction, and regression. The evidence suggests that effectiveness is governed by the match among objective, modality, and downstream task rather than by any single strategy. Contrastive objectives favor global discriminative representations suited to classification but may underrepresent localized pathology, whereas spatial-prediction, masked-modeling, and reconstruction objectives better preserve anatomical structure for segmentation and dense prediction. Critically, misaligned objectives can cause negative transfer through shortcut learning on acquisition signatures or augmentation that erases diagnostic signal rather than merely weaker gains. SSL is most beneficial in low-label regimes, but its effectiveness depends on modality-aware augmentation, pathology-preserving corruption, and clinically meaningful evaluation. We conclude with practical design guidelines and open challenges for clinically aligned SSL.

arXiv

ChangeFlow -- Latent Rectified Flow for Change Detection in Remote Sensing

Blaž Rolih, Matic Fučka, Filip Wolf, Luka Čehovin Zajc

Remote sensing change detection (RSCD) localises changes between two images of the same geographic region. Most state-of-the-art methods are trained with a per-pixel discriminative objective that classifies each spatial location independently. In this scenario, the predicted changed region is not modelled as a coherent whole, so predictions tend to be spatially fragmented. Generative modelling offers a principled solution: by learning a distribution over plausible change masks, it treats the mask as a single object and encourages global consistency. Yet existing generative RSCD methods lag behind strong discriminative baselines, held back by costly pixel-space generation and overly complex conditioning. We introduce \textbf{ChangeFlow}, which reformulates change detection as the generative synthesis of change masks in a compact latent space via rectified flow, guided by a structured yet lightweight bi-temporal conditioning signal. Changeflow yields spatially coherent predictions without sacrificing efficiency: across four binary benchmarks, SYSU, LEVIR, CLCD, and OSCD, ChangeFlow achieves an average F1 of $80.4\%$, a $1.3$-point gain over the previous best with better efficiency. It also extends to semantic change detection, setting a new state-of-the-art $65.9$ $F_{scd}$ on SECOND. Project page: https://blaz-r.github.io/changeflow_cd

arXiv

Adaptive receptive field-based spatial-frequency feature reconstruction network for fine-grained few-shot image classification

Linyue Zhang, Wenyi Zeng, Zicheng Pan, Yongsheng Gao, Changming Sun, Jun Hu, Lixian Liu, Weichuan Zhang, Tuo Wang

Feature reconstruction techniques are widely applied for few-shot fine-grained image classification (FSFGIC). Our research indicates that one of the main challenges facing existing feature-based FSFGIC methods is how to choose the size of the receptive field to extract feature descriptors (including spatial and frequency feature descriptors) from different category input images, thereby better performing the FSFGIC tasks. To address this, an adaptive receptive field-based spatial-frequency feature reconstruction network (ARF-SFR-Net) is proposed. The designed ARF-SFR-Net has the capability to adaptively determine receptive field sizes for obtaining spatial and frequency features, and effectively fuse them for reconstruction and FSFGIC tasks. The designed ARF-SFR-Net can be easily embedded into a given episodic training mechanism for end-to-end training from scratch. Extensive experiments on multiple FSFGIC benchmarks demonstrate the effectiveness and superiority of the proposed ARF-SFR-Net over state-of-the-art approaches. The code is available at: https://github.com/ICL-SUST/ARF-SFR-Net.

arXiv

Fundus Image-based Glaucoma Screening via Retinal Knowledge-Oriented Dynamic Multi-Level Feature Integration

Chi Liu, Yuzhuo Zhou, Sheng Shen, Zongyuan Ge, Fengshi Jing, Shiran Zhang, Yu Jiang, Anli Wang, Wenjian Liu, Feilong Yang, Tianqing Zhu, Xiaotong Han

While deep learning has advanced automated glaucoma screening via color fundus photography, existing purely data-driven models often overfit to confounding imaging artifacts and struggle to capture unpredictable pathological cues located beyond predefined anatomical boundaries. To address these limitations, we propose a retinal knowledge-oriented framework that synergizes dynamic multi-scale feature learning with domain-specific anatomical priors. The proposed architecture adopts a tri-branch structure to jointly model the global retinal context, the structural characteristics of the optic cup and disc, and dynamically cropped pathological regions. Specifically, we devise a Dynamic Window Mechanism (DWM) that adaptively discovers diagnostically informative patches via image-level supervision. Furthermore, we introduce a Knowledge-Enhanced Convolutional Block Attention Module (KE-CBAM) that explicitly incorporates retinal priors from a pre-trained foundation model RETFound to guide spatial attention, preventing the network from assigning spurious weights to irrelevant background noise. Extensive evaluations on the large-scale AIROGS dataset demonstrate that our method achieves a state-of-the-art AUC of $98.5\%$ and an accuracy of $94.6\%$. More importantly, the integration of anatomical priors effectively mitigates the inherent class imbalance, significantly improving the detection of referable glaucoma cases. Additional validations on the SMDG-19 benchmark further confirm its superior cross-domain generalization, indicating that our contribution provides a robust, interpretable, and scalable solution for real-world clinical glaucoma diagnosis. Our code is available at https://github.com/magiczhuo/Glaucoma-Detection

arXiv

How Affect Propagates among LLM Agents: Emergent Emotional Contagion in Crowd Simulation

Funda Durupinar

This paper studies the behavior of language models in a multi-agent crowd simulation, focusing on how affect propagates among agents that perceive and appraise one another. Each agent perceives its neighbors through visual, auditory, and tactile channels, then appraises these perceptions in light of its prompted personality profile, memory, current affective state, and situational context. Appraisal is carried out by an LLM, which updates the agent's internal affective state and selects its outward expression. The architecture contains no hand-authored mechanism for directly transferring affective state between agents; instead, inter-agent influence arises through the perception-appraisal-expression loop. The agent representation draws on the Big Five personality model and Russell's circumplex model of affect. To limit latency, low-level steering and navigation are handled by a conventional crowd simulator operating independently of the LLM-based cognitive layer. We evaluate the architecture across five scenario environments spanning alarming, joyful, and neutral situations in different spatial layouts. The results show that the system produces emotional contagion dynamics with spatial, temporal, and personality-dependent structure in sparse, small crowds. Alarm spreads from seeded agents as a traveling front, the mean alarmed fraction settles at a nonzero plateau, and the distribution of prompted personality profiles determines whether an ambiguous alarm ignites panic and whether a provocation is interpreted as anger or fear. We further evaluate the appraisal step through controlled experiments across prompt variants, sampling temperatures, and four model backends, showing that the dynamics are backend-dependent.

arXiv

Deep Label-Wise Attentive Temporal Convolutional Networks Improve Medical Coding

Muhammed Yavuz Nuzumlalı, Alexander Fabbri, Irene Li, Dragomir Radev

Medical coding is the task of assigning a set of diagnosis and procedure codes for a hospitalization using recorded notes. It requires aggregating information from different parts of the text and focus to different sections for each individual code, making it a very difficult problem even for professional human coders. We model the task as a multi-label text classification problem. To overcome the mentioned difficulties, we propose a deep neural model consisting of a multi-layer temporal convolution network (TCN) followed by label-wise attention. While multi-layer TCN helps extract a global document representation with the ability to learn relations over very long sequences, label-specific attention mechanism allows the model to focus on different aspects of the same document for each individual label. Our method achieves significantly better F-1 scores (9% increase) compared to the previous state-of-the-art model, with a remarkable increase in recall score (28% increase), which we believe is the more important metric for a clinical decision support setting.

arXiv

Multinex: Lightweight Low-light Image Enhancement via Multi-prior Retinex

Alexandru Brateanu, Tingting Mu, Codruta Ancuti, Cosmin Ancuti

Low-light image enhancement (LLIE) aims to restore natural visibility, color fidelity, and structural detail under severe illumination degradation. State-of-the-art (SOTA) LLIE techniques often rely on large models and multi-stage training, limiting practicality for edge deployment. Moreover, their dependence on a single color space introduces instability and visible exposure or color artifacts. To address these, we propose Multinex, an ultra-lightweight structured framework that integrates multiple fine-grained representations within a principled Retinex residual formulation. It decomposes an image into illumination and color prior stacks derived from distinct analytic representations, and learns to fuse these representations into luminance and reflectance adjustments required to correct exposure. By prioritizing enhancement over reconstruction and exploiting lightweight neural operations, Multinex significantly reduces computational cost, exemplified by its lightweight (45K parameters) and nano (0.7K parameters) versions. Extensive benchmarks show that all lightweight variants significantly outperform their corresponding lightweight SOTA models, and reach comparable performance to heavy models. Paper page available at https://albrateanu.github.io/multinex.

arXiv

BATON: A Multimodal Benchmark for Bidirectional Automation Transition Observation in Naturalistic Driving

Yuhang Wang, Yiyao Xu, Chaoyun Yang, Lingyao Li, Jingran Sun, Hao Zhou

Existing driving automation (DA) systems on production vehicles rely on human drivers to decide when to engage DA while requiring them to remain continuously attentive and ready to intervene. This design demands substantial situational judgment and imposes significant cognitive load, leading to steep learning curves, suboptimal user experience, and safety risks from both over-reliance and delayed takeover. Predicting when drivers hand over control to DA and when they take it back is therefore critical for designing proactive, context-aware HMI, yet existing datasets rarely capture the multimodal context, including road scene, driver state, vehicle dynamics, and route environment. To fill this gap, we introduce BATON, a large-scale naturalistic dataset capturing real-world DA usage across 127 drivers, and 136.6 hours of driving. The dataset synchronizes front-view video, in-cabin video, decoded CAN bus signals, radar-based lead-vehicle interaction, and GPS-derived route context, forming a closed-loop multimodal record around each control transition. We define three benchmark tasks: driving action understanding, handover prediction, and takeover prediction, and evaluate baselines spanning sequence models, classical classifiers, and zero-shot VLMs. Results show that visual input alone is insufficient for reliable transition prediction: front-view video captures road context but not driver state, while in-cabin video reflects driver readiness but not the external scene. Incorporating CAN and route-context signals substantially improves performance over video-only settings, indicating strong complementarity across modalities. We further find takeover events develop more gradually and benefit from longer prediction horizons, whereas handover events depend more on immediate contextual cues, revealing an asymmetry with direct implications for HMI design in assisted driving systems.

arXiv

TurPy: a physics-based and differentiable optical turbulence simulator for algorithmic development and system optimization

Joseph L. Greene, Alfred Moore, Iris Ochoa, Emily Kwan, Patrick Marano, Christopher R. Valenta

Developing optical systems for free-space applications requires simulation tools that accurately capture turbulence-induced wavefront distortions and support gradient-based optimization. Here we introduce TurPy, a GPU-accelerated, fully differentiable wave optics turbulence simulator to bridge high fidelity simulation with end-to-end optical system design. TurPy incorporates subharmonic phase screen generation, autoregressive temporal evolution, and an automated screen placement routine balancing Fourier aliasing constraints and weak-turbulence approximations into a unified, user-ready framework. Because TurPy's phase screen generation is parameterized through a media-specific power spectral density, the framework extends to atmospheric, oceanic, and biological propagation environments with minimal modification. We validate TurPy against established atmospheric turbulence theory by matching 2nd order Gaussian beam broadening and 4th order plane wave scintillation to closed-form models with 98% accuracy across weak to strong turbulence regimes, requiring only the medium's refractive index structure constant and power spectral density as inputs. To demonstrate TurPy as a gradient-based training platform, we optimize a dual-domain diffractive deep neural network (D2NN) in a two-mask dual-domain architecture to recover a Gaussian beam from a weakly turbulent path and achieving over 58% reduction in scintillation relative to an uncompensated receiver in simulation. TurPy is released as an open-source package to support synthetic data generation, turbulence-informed algorithm development, and the end-to-end design of optical platforms operating in turbulent environments.

arXiv

AutoWorld: Learning Multi-Agent Traffic Simulation with Self-Supervised World Models

Mozhgan Pourkeshavarz, Tianran Liu, Nicholas Rhinehart

Simulation with realistic traffic agents is essential for validating autonomous driving systems. Existing data-driven simulators learn agent behavior from higher-level abstractions such as 3D bounding boxes and polylines, inferred by upstream perception pipelines. These lossy abstractions discard sensory context that directly shapes agent behavior, limiting the distributional realism that simulation aims to reproduce. To address this limitation, we propose AutoWorld, a traffic simulation framework that grounds agent behavior in raw sensor observations through a self-supervised world model trained on LiDAR occupancy data. Given world model samples, AutoWorld constructs a coarse-to-fine predictive scene context as input to a multi-agent motion generation model. Furthermore, we designed a motion-aware latent supervision objective that enriches AutoWorld's latent representation of scene dynamics. To better exploit this latent space during inference, AutoWorld employs a cascaded Determinantal Point Process framework to guide diversity-aware sampling across both the world model and motion model. Experiments on the Waymo Sim Agents Challenge (WOSAC) demonstrate that AutoWorld achieves competitive performance, with larger gains in partially-observed scenarios where trajectory abstractions are most limited. We further show that grounding simulation in raw LiDAR through AutoWorld scales better with additional data than trajectory-only and LiDAR-conditioning baselines. Ablations confirm the contribution of each component.

arXiv

Towards Robust Reinforcement Learning for Small-Scale Language Model Agents

Md Rezwanul Haque, Md. Milon Islam, Fakhri Karray

The alignment of Small Language Models (SLMs) in the 70--500M parameter range using reinforcement learning is often considered unstable, though the underlying failure mechanisms have not been systematically investigated. In the State-of-the-Art (SOTA) research, fifteen (model, corpus) configurations were trained using Proximal Policy Optimization (PPO). The experiments included Pythia-70M, 160M, 410M and SmolLM2-135M, 360M on the TinyStories, CNN/DailyMail, and Wikitext-103 corpora. Three reproducible failure modes were identified in small-scale language models: silent LoRA parameter freezing in standard PEFT/TRL pipelines, numerical overflow in importance ratios when using bfloat16, and catastrophic policy collapse due to reward-model error. These issues were addressed using a merge-and-reinitialize adapter technique, float32 precision during PPO updates, and a three-layer safety mechanism comprising reward whitening, importance-ratio guarding, and weight rollback. In this paper, a capacity-headroom hypothesis is proposed, which states that PPO performance at the SLM scale depends on both a fluent supervised model ($\text{PPL}<20$) and a discriminative reward signal, rather than on the number of model parameters. The proposed system converged stably in all experiments and improved preference win rate over the SFT baseline in configurations with a fluent prior and an informative reward signal. Furthermore, it outperformed instruction-tuned baselines while requiring significantly less training data. All checkpoints, preference datasets, and training scripts are publicly released$^§$.

arXiv

DS@GT ARC at CheckThat! 2026: LLM-Based Trace Ranking and Grouped Reward Modeling for Multilingual Numerical Claim Verification

Sagnik Sinha, Shreyas Shrestha

Automated verification of numerical claims is a challenging problem, as it requires both language understanding and quantitative reasoning. This paper describes our system for CLEF 2026 CheckThat! Task 2, which focuses on ranking reasoning traces generated by large language models (LLMs) and predicting a final verdict for numerical claims in English and Arabic. We explore two approaches. The first approach fine-tunes an LLM-based verifier using LoRA to score each reasoning trace independently as a binary classification problem, and selects the final verdict using Best-of-N selection. We further experiment with adaptive sub-claim decomposition to break complex claims into simpler parts before verification. The second approach uses a lightweight TF-IDF reward model with handcrafted numeric and temporal overlap features to score traces, and aggregates scores by verdict group to determine the final prediction. For Arabic, we compare a general multilingual model against AraBERT, a language-specific model pretrained on Arabic text. Our results show that the LLM-based approach outperforms the lightweight reward model on most metrics, particularly Recall@5, while the reward-based approach shows stronger performance on the Conflicting class. Sub-claim decomposition did not improve performance, suggesting that claim splitting introduces noise rather than aiding reasoning. For Arabic, AraBERT outperforms the multilingual baseline across most metrics.

arXiv

Addressable Recall Compaction for Long Context-Window Control in AI Agents

Thang Dang, Yuma Ichikawa, Sakina Fatima, Koichi Shirahata

Long-horizon LLM agents accumulate reasoning traces, actions, and tool observations that can eventually exceed a model's fixed context window. Existing compaction methods address this limitation by discarding, summarizing, or retrieving earlier information, but they may remove task-critical details or fail to recover them reliably. We propose ARC (Addressable Recall Compaction), a context-management framework that separates archival storage from active-context presentation. ARC stores tool observations in an append-only, ID-addressable log and replaces older observations with compact citations when compaction is required. The agent can subsequently use these identifiers to request stored content without re-executing the corresponding tools or depending solely on similarity-based retrieval. We evaluate ARC using Qwen3-8B with a 16k context window and Qwen3-32B with a 32k context window. On the Needle-in-a-Haystack evaluation, ARC achieves an average exact-answer accuracy of 99.40%, compared with 88.12% for the best-performing baseline in our evaluation. ARC also reduces estimated serving time and HBM traffic under our hardware-cost model. On the LongBench-v2 Hard subset, ARC obtains an average accuracy of 29.97%, compared with 28.25% for the best-performing baseline. These results indicate that explicit, address-based recall can improve information retention and serving efficiency relative to the evaluated context-management baselines under the tested settings.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.