The study investigates whether vision–language models (VLMs) can infer human engagement from gameplay videos using the GameVibe Few‑Shot dataset across nine first‑person shooter games. Three VLMs were tested under six prompting strategies—including zero‑shot, theory‑guided prompts based on Flow, GameFlow, Self‑Determination Theory, and MDA, and retrieval‑augmented prompting—evaluating both pointwise engagement prediction and pairwise prediction of engagement change. Results show that zero‑shot predictions are weak and often do not beat simple majority‑class baselines; retrieval‑augmented prompting improves pointwise prediction in some cases, while pairwise prediction remains difficult, and theory‑guided prompts do not reliably help and may reinforce superficial shortcuts.
"whyItMatters":"The findings highlight a perception–understanding gap in current VLMs, indicating that while they can recognize visible gameplay cues, they still struggle to robustly infer human engagement across games."
By Ziyi Wang, Qizan Guo, Rishitosh Singh, Xiyang Hu
Rewind-IL is a training‑free online safeguard for generative action‑chunked imitation learning policies. It uses a zero‑shot failure detector based on Temporal Inter‑chunk Discrepancy Estimate (TIDE) and a state‑respawning mechanism that returns the robot to a verified safe intermediate state. The system builds a checkpoint library offline with a vision‑language model and monitors self‑consistency online, rewinding execution to the latest safe checkpoint when a failure is detected, thereby improving reliability in long‑horizon manipulation tasks.
By Gehan Zheng, Sanjay Seenivasan, Matthew Johnson-Roberson, Weiming Zhi
The paper investigates the edge‑of‑stability (EoS) phenomenon observed in full‑batch Adam optimization. It identifies Adam’s second‑moment adaptation as a negative‑feedback mechanism that drives the dynamics toward the stability boundary, characterized by the *active curvature*—the preconditioned curvature along the preconditioned gradient direction. The study provides rigorous characterizations across various settings, predicts *gradient reversal* near the edge (consecutive gradients pointing in nearly opposite directions), and shows that averaging iterates suppresses these oscillations, yielding smoother and lower loss curves.
By Yiman Fong, Heng Yang
The paper addresses the problem of large language models (LLMs) over-refusing to comply with benign safety-related instructions. It identifies that a small number of hypersensitive safety heads in transformer attention misfire on hard-safe prompts, causing abnormal attention entanglement and high-entropy routing conflicts that lead to refusals. The authors propose Semantic Routing Calibration (SRC), a lightweight, training‑free inference framework that dynamically suppresses these problematic heads and fuses logits to restore trustworthy reasoning while preserving intrinsic safety performance.
By Zixuan Wang, Bingjie Zhang, He Zhao, Dandan Guo
The paper introduces the Function‑Space Transformer (FST), a neural framework that learns from functions by using a spatially adaptive continuous latent representation. FST places features at anchors whose positions are predicted from input samples and refines these anchors through recursive function‑space interactions, allowing the representation to adapt to each input rather than relying on a fixed grid. Experiments on PDEBench (Burgers and Darcy flow) and ImageNet‑1K show that FST outperforms or matches state‑of‑the‑art baselines such as Perceiver IO and the Vision Transformer while using fewer parameters.
By Guorui Sang, Pedram Rooshenas
The paper introduces Uncertainty‑Normalized Margin DPO (UNM‑DPO), a method that augments direct preference optimization with strength‑dependent margins and a learned prompt scale derived from a heteroskedastic Bradley‑Terry model. Two training objectives—advantage‑only (AO) and whole‑residual (WR)—compare implicit rewards from response log‑probability ratios, with WR enabling identification of the prompt scale and a practical learning procedure. Building on WR, ULNM‑DPO‑WR further normalizes rewards by response length, and experiments on HelpSteer2/3 and AlpacaEval show higher win rates, mean rewards, and shorter responses compared to baseline DPO and SimPO.
By Sadegh Khorasani, Petrus Mikkola, Matthias Grossglauser
LEARN-TS is a method for multivariate time‑series anomaly detection that uses a frozen language model to generate two semantic representations: one that captures window‑specific observation context without exact numerical values, and another that provides a fixed, dataset‑agnostic normality prompt. These representations guide a patch‑masked reconstruction process, allowing the model to estimate normality discrepancy and produce timestamp‑level anomaly evidence. Experiments on four benchmarks show that LEARN‑TS achieves the best mean performance in 13 of 16 dataset‑metric comparisons, and ablation studies confirm the benefits of observation conditioning, joint normality alignment, and semantic references.
By Jahyeob Koo, Kio Yun, Byoungmo Koo, Jun-Geol Baek
The paper introduces SparLeak, a side‑channel attack that exploits a new GPU micro‑architectural leakage called Sparsity‑Induced Memory Access (SIMA) caused by sparse attention in large language models. By capturing SIMA traces during LLM inference, SparLeak can infer user‑query attributes from prefill‑phase traces and reconstruct autoregressive responses from decoding‑phase traces. Experiments on three LLM architectures, three sparse attention mechanisms, and three privacy‑sensitive datasets show high success rates (90.9% for attribute inference and 87.3% for response reconstruction), underscoring the need to address SIMA leakage in sparse‑attention deployments.
By Fahao Chen, Linkang Du, Jinhao Zhou, Peng Li, Zhou Su
The paper introduces ID Balancing, an Integral‑Derivative controller that improves expert load balance in extremely sparse Mixture‑of‑Experts (MoE) models. By scaling its integral term with load error and activating the derivative term only when imbalance worsens, ID Balancing achieves over 50% reduction in worst‑case backbone MaxVio and 12% reduction in training‑average backbone MinVio compared to leading baselines, while preserving language‑modeling performance across various routing settings. The method’s benefits grow with increased sparsity, making it a promising approach for scaling larger MoE models.
By Peng Jin, Zihan Qiu, Zekun Wang, Bo Zheng, Yang Xu, Tian Xie, Xiao Li, Huaqing Zhang, Haoran Lian, Rui Men, Dayiheng Liu
The paper investigates rounding strategies for low‑precision recurrent state caches in Mamba‑style and hybrid language models. It shows that a deterministic golden‑ratio Weyl dither consistently yields quantized models closer to full precision than stochastic rounding, across various models, storage formats, and long decoding horizons, without extra cost. In contrast, round‑to‑nearest can appear effective in short tests but degrades over long generations due to error accumulation.
By Snigdha Chandan Khilar
The paper investigates test‑time scaling for language models, noting that sequential methods often stall due to entering attractors—sets of answers that block further exploration. Across 27 experiments, over half of sequential trajectories hit an attractor within four iterations. Introducing a simple model‑mixing intervention reduces attractor incidence by 21.2 percentage points, expands solution coverage beyond a compute‑matched parallel baseline, and boosts recursive self‑aggregation accuracy by at least 2.2 percentage points.
By Joseph Rance, Fabio Pizzati, Juil Sock, Woody Bayliss, Marc G\'orriz Blanch, Philip Torr, Adel Bibi
The paper argues that the policy governing how tensors are rounded and reused during the backward pass—termed the backward‑state policy—is an integral part of the learning algorithm, not merely a memory or precision detail. Experiments with 390 M‑parameter models show that whether the backward pass reuses a forward’s rounded output or generates a new rounding can decisively affect training success, even when copy accuracy is high. The authors propose a method to determine, for each operator, which value should be read or substituted to preserve the correct gradient, and validate these predictions on PyTorch and Transformer Engine.
By Shuxiao Xie, Shuyang Xie, Dezhi Ran, Wei Yang, Tao Xie
The paper evaluates how language models of different sizes self-assess their confidence on question‑answering tasks across general and specialized domains. It finds that while accuracy drops for smaller models and more specialized tasks, the reliability of self‑evaluated confidence signals remains largely stable. Consequently, even less capable models can provide reasonable confidence estimates, making them suitable for resource‑constrained applications.
By Idil Kapikiran, Thomas Decker, Thomas Runkler
The paper introduces Fork-dLLM, a hybrid sampling strategy for diffusion language models that combines parallel token generation with autoregressive ordering only at uncertain fallback steps, thereby avoiding the flexibility trap that reduces generation diversity. It also presents ForkGRPO, an extension that applies the GRPO objective selectively during post‑training rollouts, preserving policy‑likelihood ratios while cutting rollout and optimization costs. Experiments show Fork-dLLM achieves pass@k scaling comparable to autoregressive sampling but 2–3× more efficient, and ForkGRPO matches or surpasses AR‑based GRPO baselines at significantly lower training expense.
By Stipe Frkovi\'c, Metod Jazbec, Christian A. Naesseth
The paper introduces a preemptive unlearning technique called gradient sealing to prevent large language models (LLMs) from acquiring forbidden capabilities during downstream fine‑tuning. It shows that existing retrospective unlearning methods fail to block hidden gradient pathways that can later enable illicit knowledge. By pushing relevant pre‑activations into the negative region, gradient sealing effectively blocks these pathways and demonstrates stronger resistance to downstream acquisition across multiple LLM families.
By Kemou Li, Qizhou Wang, Yue Wang, Fengpeng Li, Zhuan Shi, Negar Rostamzadeh, Golnoosh Farnadi, Masashi Sugiyama, Jiantao Zhou
arXiv:2609. 39892v1 Announce Type: new Abstract: Looped Transformers repeatedly apply the same set of Transformer layers, giving them a recurrent architecture for latent computation.
By Jiaju Wu, Yi Hu, Muhan Zhang
arXiv:2609. 40148v1 Announce Type: new Abstract: Power-law learning curves are often treated as fixed properties of a model and its data, although learning-rate and batch-size schedules can change the observed loss.
By Yichen Wang, Fanghui Liu, Yudong Chen
The study shows that previously reported gains in decoding spoken words from non‑invasive brain recordings can largely be replicated without any brain data, due to a timing shortcut where overlapping windows reveal word durations. By processing each window independently rather than jointly, the authors eliminate this shortcut, enabling the neural network to learn genuine brain‑based word information. This change improves performance and makes two existing strategies—aggregating predictions from multiple neural responses and incorporating a pretrained language model—significantly more effective, achieving a word error rate of 36.6% on a perceived speech benchmark.
By Dulhan Jayalath, Oiwi Parker Jones
The paper introduces VirusCascade, a black‑box attack that exploits the collaborative reflection mechanism in LLM‑powered recommender agents. By leveraging reflective persistence and cross‑agent propagation, the attack promotes a target item through semantic rationalisation and structural positioning, achieving state‑of‑the‑art targeted exposure on four real‑world datasets. The study demonstrates that existing static‑pipeline attacks cannot exploit this recurrent, multi‑agent amplification pathway.
By Yurong Hao, Wen Zhou, Guowei Guan, Tiantong Wu, Fuyao Zhang, Wei Yang Bryan Lim
The paper introduces the Posterior Concentration Phenomenon (PCP), a length‑dependent failure mode where probability‑based rewards collapse to a narrow interval for long reasoning traces, destabilizing verifier‑free reinforcement learning. To address this, the authors propose RLCPR, a framework that uses uncertainty‑aware data sampling and concentration‑aware regularization to mitigate PCP, improving token efficiency and outperforming state‑of‑the‑art baselines on multiple reasoning benchmarks.
By Shiu-Hong Kao, Yubo Zhao, Zhenyu Tian, Pengzhan Sun, Yicong Li, Angela Yao