Research

• Interface lock-in in multimodal spatial reasoning1 paper raises it · 1 addresses it

  Specializing a model on one spatial task (e.g., side detection) degrades performance on others (e.g., distance estimation), limiting general spatial intelligence.

  • ChronoReplay: One Answer Interface Is Not Enough for Streaming Spatial Intelligenceraises · addresses
• Lack of unified evaluation for work agents1 paper raises it · 1 addresses it

  Existing benchmarks do not reliably measure how agents execute complex, stateful, or open-ended digital work, making progress hard to track.

  • Work Agents: Systems, Capabilities, and Optimizationraises · addresses
• Memory tax in long-horizon agents1 paper raises it · 1 addresses it

  Dedicated memory systems often underperform naive retrieval, as reorganizing information can obscure evidence and reduce net value.

  • Memory Tax: Dissecting the Net Value of Agent Memoryraises · addresses
• Credit assignment in diffusion language models1 paper raises it · 2 address it

  Iterative denoising in DLMs makes it hard to attribute rewards to individual tokens, limiting effective reinforcement learning.

  • DTCA: Denoising Token Credit Assignment for Reinforcement Learning in Diffusion Language Modelsraises · addresses
  • Multiscale Policy Optimization for Long-Horizon LLM Agentsaddresses
• Grounded significance testing for temporal hypotheses1 paper raises it · 1 addresses it

  Language models assert temporal patterns without rigorous statistical validation, risking incorrect conclusions in high-stakes domains like healthcare.

  • Testimony from Timelines: Evidence-Grounded Significance Testing of Temporal Clinical Hypothesesraises · addresses

Named by an AI from these papers' abstracts.

• Character recognition OCR11+10.0%
• AI tutors14+7.7%
• Channel state information120.0%
• Supervised & diffusion models93−8.8%
• Electrocardiogram ecg34−15.0%
• Stochastic gradient descent24−22.6%
• Quantum machine learning14−26.3%
• Machine learned interatomic33−31.2%

• arXiv cs.CLOct 6▲ 27

  UNREAL: Unifying Retrieval and Long-Context with a Single Model

  Edan Kinderman, Elad Hoffer, Yochai Blau, Brian Chmiel, Ron Banner +2

  Long-context inference and Retrieval-Augmented Generation (RAG) handle evidence selection at vastly different scales, from a single long prompt to an entire corpus. We ask whether a single model-internal mechanism can select evidence across this range. We introduce UNifying REtrieval And Long-Context with a Single Model (UNREAL), a model-native evidence selection framework to span corpus retrieval and long-context inference. UNREAL encodes chunks and derives retrieval queries directly from the frozen LLM's internal representations. It adds fewer than 500K trainable parameters and leaves the backbone unchanged. On a 3B-token, 21M-chunk Wikipedia index, all four dense and hybrid UNREAL backbones outperform state-of-the-art retriever-reranker systems. The best model raises recall from 49.1% to 73.2% on HotpotQA, from 31.7% to 60.1% on 2WikiMultiHopQA, and from 8.8% to 14.4% on MuSiQue. Applied to long-context tasks, the same selection mechanism removes distractors before generation, raising NoLiMa accuracy from 1.0% to 24.83% at its maximum context length of 128K tokens, and LV-Eval's F1 score from 49.97% to 54.66% at 256K. UNREAL also reduces FLOPs and time-to-first-token relative…

  PDF

  Summarize this
• arXiv cs.CLOct 6

  Agentic AutoRAG: RAG Pipeline Optimization through Reasoning-Driven Agents

  Lasse B. Strand, Robert Jakob, Kevin O'Sullivan, Markus Kreft

  Retrieval-augmented generation (RAG) is a widely used approach for grounding large language models (LLMs) in external knowledge. However, configuring a pipeline is an expensive hyperparameter optimization problem over many interacting choices, from chunking and embedding model to reranking and generation. Existing optimizers, from greedy search to Bayesian optimization, reduce each trial to an aggregate score and search without modeling why a configuration performed as it did, even though the retrieved chunks already provide evidence about whether each failure occurred during retrieval or after it. We introduce Agentic AutoRAG, an LLM-agent optimizer for multi-objective RAG hyperparameter optimization with retrieval-versus-generation failure attribution. It proposes configurations scored on a frozen exam from the corpus: after each trial a Diagnoser attributes each failed question to retrieval or generation, and a Proposer, grounded in a knowledge base of model rankings and pricing, selects the next configuration, weighing accuracy against cost to trace a Pareto frontier. On three multi-hop QA benchmarks it reaches higher LLM-judge accuracy than every baseline we compare, and…

  PDF

  Summarize this
• arXiv cs.CLOct 6▲ 213

  Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability

  Bingxi Hou, Guochao Jiang, Guofeng Quan, Weiqing Li, Wenfeng Feng +2

  On-Policy Distillation (OPD) trains a student on its own generations using teacher feedback. With different tokenizers, comparing teacher and student predictions requires alignment at both sequence and vocabulary levels. In this paper, we examine whether expanding this alignment coverage improves learning. Across three heterogeneous teacher--student pairs on mathematical reasoning and code generation, strict 1:1 groups already cover most student-generated tokens despite substantial vocabulary mismatch. On responses sampled from the students before distillation, the shared vocabulary retains nearly all teacher and student probability mass at strictly aligned positions on average. Restricting reverse KL to a student-selected top-16 subset of the shared vocabulary at each strict position achieves accuracy comparable to full shared-vocabulary OPD, outperforming the evaluated cross-tokenizer baselines. Adding mean squared error supervision on span log-probabilities in mismatch groups gives complete supervision coverage, yet reduces accuracy. At checkpoints from training with only the strict loss, the span gradients show weak or negative directional agreement with the strict gradients…

  PDF

  Summarize this
• arXiv cs.AIOct 6

  AssemState: Manual and Physical-State-Guided Reasoning for Zero-shot Furniture Assembly

  Zhiyuan Qi, Jierui Li, Yifan Shen, Cheng Qian, Jiateng Liu

  Multimodal large language models (MLLMs) have made significant progress in visual understanding, but precise 3D spatial reasoning integrated with physical environment remains difficult. Furniture assembly requires not only recovering step-level operations from diagrammatic manuals, but also translating semantic attachment relations into 6D pose updates that enable parts to physically interact with the environment and previously assembled components. To study this problem, we propose AssemState, a zero-shot framework for manual and physical-state-guided furniture assembly. It firstly employs anchor-guided boundary assembly states to decompose manual pages into single-part operations and recover an assembly-tree. Then, it uses iterative after-state feedback refinement to guide successive (SE(3)) updates and corrections, and validates their physical plausibility through simulation-based release tests. Experiments show that compared with the strongest prior baseline, AssemState improves F1 from 38.58\% to 62.80\% and Tree Exact Match from 28.24\% to 53.92\% for assembly-tree recovery. On 243 independently evaluated part-level operations, our proposed iterative refinement improves…

  PDF

  Summarize this
• arXiv cs.CVOct 6

  HuC-VideoMAE: Human-Centric Video Masked Autoencoding from synthetic data

  Ricardo Pizarro, Roberto Valle, José M. Buenaposada, Luis M. Bergasa, Luis Baumela

  Modern action recognition models rely on video transformers pretrained on massive collections of web-crawled videos, such as Kinetics-700. However, the use of such data raises ethical concerns, as subjects' consent is typically not obtained. Recent high-quality synthetic video datasets generated from motion-capture data, such as BEDLAM2.0, offer a promising ethical alternative. In this work, we investigate self-supervised pretraining of video transformers on synthetic human-motion datasets. We first show that directly applying the standard VideoMAE masking strategy leads to substantially worse performance than pretraining on Kinetics. To address this limitation, we propose a human-centric masking scheme that leverages body keypoints and person bounding box regions. Our approach encourages the model to focus on the structure and dynamics of human motion during pretraining. Experiments on NTU RGB+D and Toyota-Smarthome demonstrate that our method significantly outperforms standard VideoMAE pretraining on synthetic data, closing 49% of the gap to Kinetics pretraining on NTU RGB+D cross-view-subject without using a single real frame during pretraining. To promote the use of ethical…

  PDF

  Summarize this
• arXiv cs.AIOct 6▲ 10

  EMHO: EMbodied Agent Harness Optimization via Experience Traces

  Hyun Jung Lee, Jungtaek Kim, Jongwon Jeong, Tae-Eui Kam, Donghyun Kim +1

  Improving embodied agents often focuses on optimizing the underlying model through training, while the surrounding agent harness that controls planning, context, and tool use is typically engineered. We ask whether this harness can instead improve itself directly from experience traces under sparse environmental feedback. We propose EMbodied Agent Harness Optimization (EMHO), a self-evolving framework that keeps the embodied model frozen and iteratively revises its harness by analyzing execution trajectories and prior harness history. EMHO optimizes beyond skills or recovery prompts, modifying how the agent monitors progress, uses vision tools, grounds observations, and responds to failures. To support multiple subtasks with a single harness, we introduce EMHO-Merge, which addresses trade-offs in jointly optimizing a single shared harness across subtasks by using episode-level gains and losses to guide evidence-supported refinement of when and how revised behaviors are applied. We evaluate EMHO on EmbodiedBench across navigation and manipulation tasks, and EMHO consistently improves task success for both Qwen 9B and 27B models. Qualitative analysis shows that EMHO goes beyond…

  PDF

  Summarize this
• arXiv cs.DCOct 6▲ 29

  NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale

  Songlin Jiang, Zhiyu Li, Terry Kong, Yu Yao, Youngeun Kwon +3

  Agentic reinforcement learning (RL) disaggregates training from rollout, so each policy update must reach the rollout clusters before the next batch. Transferring a full 1T checkpoint for such weight synchronization (refit) takes 87.5 min between two AWS regions. Measurements of BF16 training show that about 1% of weights change their stored values per step. Recent systems exploit this sparsity but fall short on placement, exactness, or efficiency: they reimplement placement rules, assemble full tensors, rebuild values arithmetically, or use a cross-cluster collective, and none fully recovers from mid-refit failures. We present NeMo-DCR (Delta-Compressed Refit), which sends only changes yet is bit-exact: receivers obtain the same parameter and buffer bits as a dense refit. For placement, fixed affine mappings project changes from training shards into the checkpoint's canonical coordinates, residual conversion covers the other changes, and the serving runtime's native loader places all changes in receiver storage. For exactness, compressible XOR masks carry affine changes whose projection and loader preserve stored bits, and overwrites carry the others. Receivers apply both in…

  Code
• arXiv cs.LGOct 6

  Symmetry-Aware Feature Learning: A Polynomial Separation for Multi-Index Models

  Jivan Waber, Vanessa Piccolo, Yatin Dandi, Florent Krzakala

  We establish a polynomial sample complexity separation between symmetry-aware and symmetry-agnostic feature learning. We study growing-rank multi-index models with high-dimensional Gaussian covariates in $\mathbb{R}^d$ and $r=Θ(d^δ)$ teacher directions forming a cyclic symmetry orbit, where $0<δ<1/2$. We compare three ways of exploiting this structure: architectural weight sharing, data augmentation over the full symmetry group, and learning without access to the symmetry. In particular, we analyze a symmetry-tied convolutional network, an untied network, and the same untied network trained with full-group data augmentation, using spherical online SGD with correlation loss. For a class of polynomial links with information exponent $p\ge3$, we prove matching sample complexity bounds up to logarithmic factors: the tied and augmented learners achieve weak directional recovery in $\widetildeΘ(d^{p-1})$ samples, whereas the symmetry-agnostic learner requires $\widetildeΘ(rd^{p-1})$. For the pure quadratic Hermite link, the same separation holds for weak recovery of the teacher subspace, with sample complexities $\widetildeΘ(d)$ and $\widetildeΘ(rd)$, respectively. Thus, full-group…

  PDF

  Summarize this
• arXiv cs.CVOct 6

  Deformable CT-US Registration via Anatomy-Aware Implicit Neural Representations

  Agnieszka Lach, Magdalena Wysocki, Feng Li, Mohammad Farid Azampour, Benjamin D. Killeen +5

  Slice-to-volume registration between ultrasound (US) and preoperative computed tomography (CT) imaging would enhance many minimally invasive interventions, for example by locating soft tissue structures intra-operatively that are discernible in CT. While optical tracking enables initial rigid registration, contact from the probe induces soft tissue deformations that inhibit accurate alignment. In this work, we introduce a deformable CT-ultrasound registration framework that incorporates anatomical priors derived from CT to improve registration under deformation. Rigid registration is first established using a robot-assisted optical tracking system, after which a deformable transformation is estimated using a sinusoidal implicit neural representation (SIREN) optimized per frame. Tissue stiffness is approximated from CT-based HU values and used as spatially varying regularization, suppressing deformation in rigid structures such as bone while allowing more flexibility in soft tissue. Two additional constraints capture the physics of probe contact: a contact-zone displacement prior that drives the displacement field to compress tissue below the probe face, and a fan-geometry…

  PDF

  Summarize this
• arXiv cs.CVOct 6

  From the Drosophila Visual Connectome to General-Purpose Computer Vision

  Zongyu Li, Akito Yamauchi, Huaizhi Liu, Vishwanatha Rao, Jia Guo +2

  Biological connectomes encode structured solutions to visual computation that may provide reusable inductive biases for artificial vision. We develop ConnectomeX around FlyVision, a trainable architecture that preserves parallel ON/OFF processing, recurrent computation and population-level graph interaction while scaling model capacity across tasks. FlyVision reached 99.34% accuracy on MNIST with 80,608 parameters and 78.03% on CIFAR-10 with 81,408 parameters. On ImageNet-1K, FlyVision Base and Large reached 60.79% and 66.25% top-1 accuracy with 1.8 and 3.7 million parameters, while a Large local-k7 model with a learned low-frequency branch reached 66.53%, compared with 69.25% for ResNet18 with 11.7 million parameters. On a 22-class skin-disease benchmark, FlyVision Large achieved 63.78% accuracy and 95.28% macro-AUROC with 2.99 million parameters. In four-class chest radiography, ImageNet-pretrained FlyVision Base and Large reached 92.60% and 92.76% accuracy with 1.33 and 2.97 million parameters, compared with 91.56% for ImageNet-pretrained ResNet18 with 11.18 million. BrainAGE extends FlyVision to volumetric T1-weighted MRI by applying a shared ImageNet-pretrained FlyVision…

  PDF

  Summarize this
• arXiv cs.CVOct 6

  Ariadne's Thread of LipSync: Unraveling Forgeries via Inconsistency between Lip Motions and Head Poses

  Tianyi She, Jiawei Liu, Weifeng Liu, Hanqing Zhao, Weiming Zhang +1

  Recent advances in LipSync generation technology have led to the creation of highly realistic videos, posing severe societal risks. However, existing defense strategies struggle against LipSync forgeries, as advanced LipSync generation methods not only achieve better lip synchronization but also eliminate visual artifacts. An important reason is that they overlook an inherent biological coupling between lip movements and head poses in natural speech videos. In this paper, we propose LipDA, a novel framework for joint LipSync Detection and Attribution, which takes advantage of the inconsistency between head and lip. For detection, the framework learns to quantify this discrepancy by contrasting lip and pose features from authentic versus forged videos. For attribution, our method is designed to capture the unique temporal dynamics and audio-visual synchronization patterns that act as the fingerprint of models, enabling source tracing. We conduct extensive experiments on two challenging LipSync datasets as well as our own proposed large-scale and multi-generator dataset. LipDA achieves over 97\% AUC in detection and 97.5\% accuracy in model attribution, significantly outperforming…

  PDF

  Summarize this
• arXiv cs.CVOct 6

  Image Bitstream Fine-grained Understanding for Privacy-Friendly AIoT

  Zhen Yu, Wenyang Liu, Kejun Wu, Chengwang Xiao, Renjie Qiao +1

  Image Bitstream Fine-grained Understanding (IBFU) aims to directly perform fine-grained classification and semantic description generation from encoded image byte sequences. In contrast to conventional pixel-domain visual understanding, IBFU conducts semantic analysis without fully decoding images into the pixel domain. Since pixel-level visual content is not explicitly reconstructed during inference, this paradigm reduces visual exposure within the processing pipeline and suits privacy-friendly Artificial Intelligence of Things (AIoT) applications. In this paper, we propose Bitstream Fine-grained Generator (BFG), a novel foundation model tailored for IBFU. BFG consists of two main components: a Bitstream Semantic Encoder (BSeE) and a Fine-grained Semantic Generator (FSeG). BSeE directly models semantic representations from encoded image bitstreams without explicit pixel reconstruction, while FSeG transforms the extracted bitstream semantics into detailed natural-language descriptions through autoregressive generation. To train BFG and comprehensively evaluate IBFU in practical AIoT scenarios, where image bitstreams may suffer corruption during transmission and storage, we…

  PDF

  Summarize this
• arXiv cs.CLOct 6

  Knowing When Not to Answer: Cross-Domain and Multi-Turn Generalization of Latent Underspecification Signals

  Jerzy Kamiński, Ilya Galyukshev, Artem Kuznetsov, Danil Fedorov, Kirill Redko +4

  Large language models routinely answer questions that cannot be answered from the information given, and in dialogue they answer before enough has been said. Unanswerability is linearly decodable from hidden states, but it is unclear which of its forms share a representation and whether the signal is useful in dialogue. We contribute a turn-labeled multi-turn benchmark (423 conversations, 1,661 labeled turn-states) and an evaluation harness with a simulated user who answers clarifying questions, and use them with six datasets and six open-weight LLMs to test how far probes for unanswerability carry. Probes transfer robustly between datasets that share a ground of unanswerability: missing information in math (AUROC 0.77-0.97) and in a passage (SQuAD 2.0<->MuSiQue, 0.77-0.90). Probes for epistemic "known-unknowns" transfer poorly to math, but this separation weakens under lexical controls and changes with layer and coordinate system, so it remains unresolved. Single-turn probes fail zero-shot to detect when a conversation becomes answerable; in-structure probes recover it, but no better than a bag-of-words classifier. A gate on the calibrated probe, with no model fine-tuning, fires…

  PDF

  Summarize this
• arXiv cs.CVOct 6

  Decoy and disclosure radii of invariant shape descriptors

  Tanush Shaska, Lubjana Beshaj

  A recognizer that compares rotation-invariant descriptors sees a surface only up to the fiber of the descriptor. We measure this fiber by its radius in the orbit distance from the enrolled surface. A large radius admits decoys, that is, distant shapes that pass the matcher. A small radius discloses the enrolled shape to anyone who captures the stored value. For star-shaped surfaces truncated to spherical harmonics of degree at most $L$, with $n$ coefficients, a descriptor of generic rank $r$ has generic fibers of dimension $n-3-r$ modulo rotations. The standard pool of band powers, even bispectra, and three invariants of the degree-three band therefore admits decoy families of dimension $5$, $13$, $20$ at $L=4,6,8$. Its rank first reaches $n-3$ at $L=16$, and a mirror decoy remains at every $L$. The odd bispectra remove the mirror decoy generically for $L \geq 4$. Yet at fixed mean radius the same pool determines the enclosed volume exactly, and it does not determine whether a surface meets a clearance requirement. We certify two cases by exact and interval arithmetic. At $L=6$ a decoy matches all $32$ invariants to relative precision $2 \cdot 10^{-18}$ at orbit distance at least…

  PDF

  Summarize this
• arXiv cs.SEOct 6

  Learning from Failures: A Failure-Driven Prompt Refinement for LLM-Based Vulnerability Analysis

  Mandana Ghadamian, David Mohaisen

  Large Language Models have emerged as promising tools for software vulnerability analysis, but their effectiveness depends heavily on prompt design. Existing research primarily compares prompting strategies using aggregate performance metrics, providing limited insight into why models fail or how prompts can be improved systematically. We propose Failure-Driven Prompt Refinement (FDPR), a methodology that analyzes recurring model failures to guide evidence-based prompt refinement. Using the Damn Vulnerable Java Application (DVJA), we identify recurring failure modes, including false positives, false negatives, unsupported reasoning, and CWE misclassification, and translate them into targeted prompt refinements. We then evaluate the resulting prompt on the Juliet Test Suite and perform cross-model validation to assess generalizability. The results show that failure-driven refinement improves the reliability of LLM-based vulnerability analysis while yielding reusable prompt design principles. More broadly, this work demonstrates that recurring model failures provide a principled foundation for prompt engineering, enabling the systematic development of more reliable LLM-based…

  PDF

  Summarize this
• arXiv cs.LGOct 6

  SSR: Sparse Segment Reduction for Ternary GEMM Acceleration

  Adeline Pittet, Shien Zhu, Valérie Verdan, Gustavo Alonso

  Large Language Models (LLMs) require substantial computational resources, limiting their deployment on resource-constrained hardware. Ternary LLMs mitigate these demands through weight quantization via ternary values, achieving significant compression often with 50-90% sparsity. However, existing approaches have limitations: methods optimized for ternary weights, such as BitNet, redundant segment reduction (RSR), and its improved version RSR++, do not exploit sparsity structures, while conventional sparse formats neglect ternary characteristics, foregoing dual optimization opportunities. In this paper, we introduce Sparse Segment Reduction (SSR), a ternary matrix multiplication method designed to accelerate the inference of ternary LLMs and general Ternary Weight Networks (TWNs). SSR has a dedicated optimized ternary data format and an algorithm that systematically exploits sparsity patterns through computation trees that scale with the sparsity. SSR provides theoretical gains with asymptotically faster inference than RSR++ for sparsity above 50%, while practical evaluations reveal performance improvements across all sparsity levels. Evaluation results show that SSR achieves…

  PDF

  Summarize this
• arXiv cs.LGOct 6

  VETTA: Coordinating Turn- and Token-Level Credit Assignment for Multi-Turn LLM Agents

  Jiaju Chen, Min Yang, Jinghua Piao, Xiaochong Lan, Xu Xia +2

  Multi-turn LLM agents often receive sparse task feedback across several interactions, while generating each response token by token. This creates two related credit-assignment questions: which responses helped achieve the outcome, and which generation decisions mattered within each response? Existing methods typically focus on only one level: turn-level methods evaluate complete responses but do not distinguish the decisions within them; token-level methods can propagate feedback across turns but do not explicitly model credit for each response. These complementary limitations motivate learning credit at both levels and coordinating it in a single policy update. We introduce VETTA, a credit assignment method that jointly learns turn- and token-level values through separate heads on a shared lightweight critic. VETTA computes advantages along both temporal sequences and combines each turn advantage with a within-response-centered token residual for PPO updates. Furthermore, to reduce value-learning cost, the critic retains only early Transformer blocks from the pretrained checkpoint used to initialize the actor. On two challenging agent benchmarks, ALFWorld and WebShop, VETTA…

  PDF

  Summarize this
• arXiv cs.CVOct 6

  GeoPID: Decomposing and Steering Visual Information in Vision-Language Models

  Seulgi Kim, Zhixiong Zhang, Xinwei Zhang, Jie Ling, Ronn Shaw

  While recent vision-language models (VLMs) have shown outstanding performance across diverse applications, they tend to under-use visual information and over-rely on textual context. In this work, we propose \textsc{GeoPID}, a training-free framework that analyzes multimodal information within VLMs from a geometric perspective. \textsc{GeoPID} decomposes information into Redundant, Modality-Unique, and Synergistic components through the geometric relationships between visual and textual representation subspaces. Through an extensive analysis across 22 VLMs and 14 benchmarks, we confirm that correct predictions exhibit stronger vision-unique components when questions strongly require visual grounding. Building on this geometric analysis, we introduce a targeted intervention technique that selectively amplifies visual representations along the vision-unique subspace during inference. As a result, visual grounding capabilities were enhanced without any additional model parameter updates, achieving an average relative accuracy gain of 7.63\%.

  PDF

  Summarize this
• arXiv cs.LGOct 6

  Atom-JEPA: Joint-Embedding Predictive Architecture for 3D Atomistic Systems

  Kasper Helverskov Petersen, Rasmus Hannibal Tirsgaard, François R J Cornet, Mikkel Jordahn, Mikkel N. Schmidt

  Large-scale self-supervised pretraining has reshaped modern machine learning, substantially advancing the ability of language and vision models to generalize across downstream tasks. While deep learning has driven considerable progress in modeling atomistic systems in recent years, self-supervised pretraining in this domain has not yet achieved comparable downstream generalization. To address this, we introduce Atom-JEPA, a self-supervised pretraining framework that learns latent representations from unlabeled 3D structures through complementary atom-level and substructure-level objectives inspired by joint-embedding predictive architectures. We pretrain Atom-JEPA on large-scale molecular and crystalline datasets and evaluate its transfer performance by fine-tuning on a diverse set of downstream property prediction tasks. Atom-JEPA achieves state-of-the-art performance on molecular ADMET and quantum-chemical property prediction tasks, and is highly competitive in predicting the physical properties of crystalline materials. These results demonstrate the potential of latent-space predictive pretraining to support broad downstream generalization from structural data alone. Code and…

  PDF

  Summarize this
• arXiv cs.CVOct 6

  UP-MOPD: Update Projection in Multi-Teacher On-Policy Distillation

  Taojie Zhu, Jing Jin, Yuan Xia, Chenyang Ding, Qunshan He +6

  On-policy distillation from multiple teachers combines expertise from different domains in a single student, but conflicting gradients can hinder this integration. Gradient corrections directly constrain parameter updates under plain SGD. With optimizers such as AdamW, however, momentum, adaptive scaling, and weight decay can turn a corrected gradient into an update that increases a domain loss to first order. To address this gap, we propose Update Projection for Multi-Teacher On-Policy Distillation (UP-MOPD). UP-MOPD lets the original mixed gradient update the optimizer state and generate a candidate displacement, then projects only violating candidates before they are committed to the parameters. The projection gives the unique feasible update closest to the candidate in Euclidean distance. In experiments combining medical and general domains, UP-MOPD improves IFEval-loose accuracy late in training by 2.96 points over vanilla M-OPD. It achieves an average score of 60.03 across eight metrics, compared with 59.00 for gradient projection and 59.15 for update rejection. On a public benchmark covering mathematics, code, and instruction following, it achieves the best average across…

  PDF

  Summarize this