DeformTrace: A Deformable State Space Model with Relay Tokens for Temporal Forgery Localization

Temporal Forgery Localization (TFL) aims to precisely identify manipulated segments in video and audio, offering strong interpretability for security and forensics. While recent State Space Models (SSMs) show promise in precise temporal reasoning, their use in TFL is hindered by ambiguous boundaries, sparse forgeries, and limited long-range modeling. We propose DeformTrace, which enhances SSMs with deformable dynamics and relay mechanisms to address these challenges. Specifically, Deformable Self-SSM (DS-SSM) introduces dynamic receptive fields into SSMs for precise temporal localization. To further enhance its capacity for temporal reasoning and mitigate long-range decay, a Relay Token Mechanism is integrated into DS-SSM. Besides, Deformable Cross-SSM (DC-SSM) partitions the global state space into query-specific subspaces, reducing non-forgery information accumulation and boosting sensitivity to sparse forgeries. These components are integrated into a hybrid architecture that combines the global modeling of Transformers with the efficiency of SSMs. Extensive experiments show that DeformTrace achieves state-of-the-art performance with fewer parameters, faster inference, and stronger robustness.

Interpretable Pre-Release Baseball Pitch Type Anticipation from Broadcast 3D Kinematics

How much can a pitcher's body reveal about the upcoming pitch? We study this question at scale by classifying eight pitch types from monocular 3D pose sequences, without access to ball-flight data. Our pipeline chains a diffusion-based 3D pose backbone with automatic pitching-event detection, groundtruth-validated biomechanical feature extraction, and gradient-boosted classification over 229 kinematic features. Evaluated on 119,561 professional pitches, the largest such benchmark to date, we achieve 80.4\% accuracy using body kinematics alone. A systematic importance analysis reveals that upper-body mechanics contribute 64.9\% of the predictive signal versus 35.1\% for the lower body, with wrist position (14.8\%) and trunk lateral tilt emerging as the most informative joint group and biomechanical feature, respectively. We further show that grip-defined variants (four-seam vs.\ two-seam fastball) are not separable from pose, establishing an empirical ceiling near 80\% and delineating where kinematic information ends and ball-flight information begins.

SEA-TS: Self-Evolving Agent for Autonomous Code Generation of Time Series Forecasting Algorithms

Accurate time series forecasting underpins decision-making across domains, yet conventional ML development suffers from data scarcity in new deployments, poor adaptability under distribution shift, and diminishing returns from manual iteration. We propose Self-Evolving Agent for Time Series Algorithms (SEA-TS), a framework that autonomously generates, validates, and optimizes forecasting code via an iterative self-evolution loop. Our framework introduces three key innovations: (1) Metric-Advantage Monte Carlo Tree Search (MA-MCTS), which replaces fixed rewards with a normalized advantage score for discriminative search guidance; (2) Code Review with running prompt refinement, where each executed solution undergoes automated review followed by prompt updates that encode corrective patterns, preventing recurrence of similar errors; and (3) Global Steerable Reasoning, which compares each node against global best and worst solutions, enabling cross-trajectory knowledge transfer. We adopt a MAP-Elites archive for architectural diversity. On the public Solar-Energy benchmark, SEA-TS generated code achieves a 40% MAE reduction relative to TimeMixer, surpassing state-of-the-art methods. On proprietary datasets, SEA-TS generated code reduces WAPE by 8.6% on solar PV forecasting and 7.7% on residential load forecasting compared to human-engineered baselines, and achieves 26.17% MAPE on load forecasting versus 29.34% by TimeMixer. Notably, the evolved models discover novel architectural patterns--including physics-informed monotonic decay heads encoding solar irradiance constraints, per-station learned diurnal cycle profiles, and learnable hourly bias correction--demonstrating that autonomous ML engineering can generate genuinely novel algorithmic ideas beyond manual design.

Osmosis Distillation: Model Hijacking with the Fewest Samples

Transfer learning is devised to leverage knowledge from pre-trained models to solve new tasks with limited data and computational resources. Meanwhile, dataset distillation has emerged to synthesize a compact dataset that preserves critical information from the original large dataset. Therefore, a combination of transfer learning and dataset distillation offers promising performance in evaluations. However, a non-negligible security threat remains undiscovered in transfer learning using synthetic datasets generated by dataset distillation methods, where an adversary can perform a model hijacking attack with only a few poisoned samples in the synthetic dataset. To reveal this threat, we propose Osmosis Distillation (OD) attack, a novel model hijacking strategy that targets deep learning models using the fewest samples. Comprehensive evaluations on various datasets demonstrate that the OD attack attains high attack success rates in hidden tasks while preserving high model utility in original tasks. Furthermore, the distilled osmosis set enables model hijacking across diverse model architectures, allowing model hijacking in transfer learning with considerable attack performance and model utility. We argue that awareness of using third-party synthetic datasets in transfer learning must be raised.

FireBench: Evaluating Instruction Following in Enterprise and API-Driven LLM Applications

Instruction following is critical for LLMs deployed in enterprise and API-driven settings, where strict adherence to output formats, content constraints, and procedural requirements is essential for enabling reliable LLM-assisted workflows. However, existing instruction following benchmarks predominantly evaluate natural language generation constraints that reflect the needs of chat assistants rather than enterprise users. To bridge this gap, we introduce FireBench, an LLM instruction following benchmark grounded in real-world enterprise and API usage patterns. FireBench evaluates six core capability dimensions across diverse applications including information extraction, customer support, and coding agents, comprising over 2,400 samples. We evaluate 11 LLMs and present key findings on their instruction following behavior in enterprise scenarios. We open-source FireBench at fire-bench.com to help users assess model suitability, support model developers in diagnosing performance, and invite community contributions.

SinhaLegal: A Benchmark Corpus for Information Extraction and Analysis in Sinhala Legislative Texts

SinhaLegal introduces a Sinhala legislative text corpus containing approximately 2 million words across 1,206 legal documents. The dataset includes two types of legal documents: 1,065 Acts dated from 1981 to 2014 and 141 Bills from 2010 to 2014, which were systematically collected from official sources. The texts were extracted using OCR with Google Document AI, followed by extensive post-processing and manual cleaning to ensure high-quality, machine-readable content, along with dedicated metadata files for each document. A comprehensive evaluation was conducted, including corpus statistics, lexical diversity, word frequency analysis, named entity recognition, and topic modelling, demonstrating the structured and domain-specific nature of the corpus. Additionally, perplexity analysis using both large and small language models was performed to assess how effectively language models respond to domain-specific texts. The SinhaLegal corpus represents a vital resource designed to support NLP tasks such as summarisation, information extraction, and analysis, thereby bridging a critical gap in Sinhala legal research.

Why Is RLHF Alignment Shallow? A Gradient Analysis

Why is safety alignment in LLMs shallow? We prove that gradient-based alignment inherently concentrates on positions where harm is decided and vanishes beyond. Using a martingale decomposition of sequence-level harm, we derive an exact characterization of alignment gradients. The gradient at position $t$ equals the covariance between the conditional expected harm and the score function. This implies that positions beyond the harm horizon where the output's harmfulness is already determined receive zero gradient signal during training. This explains empirical observations that KL divergence between aligned and base models concentrates on early tokens. Consequently, standard alignment objectives cannot produce deep alignment, regardless of optimization quality. We introduce the concept of harm information $I_t$, which quantifies each position's influence on harm, and prove that equilibrium KL divergence tracks this quantity. Finally, we derive an objective based on recovery penalties that creates gradient signal at all positions, providing theoretical grounding for empirically successful data augmentation techniques.

Beyond Text: Aligning Vision and Language for Multimodal E-Commerce Retrieval

Modern e-commerce search is inherently multimodal: customers make purchase decisions by jointly considering product text and visual informations. However, most industrial retrieval and ranking systems primarily rely on textual information, underutilizing the rich visual signals available in product images. In this work, we study unified text-image fusion for two-tower retrieval models in the e-commerce domain. We demonstrate that domain-specific fine-tuning and two stage alignment between query with product text and image modalities are both crucial for effective multimodal retrieval. Building on these insights, we propose a noval modality fusion network to fuse image and text information and capture cross-modal complementary information. Experiments on large-scale e-commerce datasets validate the effectiveness of the proposed approach.

Multilevel Training for Kolmogorov Arnold Networks

Algorithmic speedup of training common neural architectures is made difficult by the lack of structure guaranteed by the function compositions inherent to such networks. In contrast to multilayer perceptrons (MLPs), Kolmogorov-Arnold networks (KANs) provide more structure by expanding learned activations in a specified basis. This paper exploits this structure to develop practical algorithms and theoretical insights, yielding training speedup via multilevel training for KANs. To do so, we first establish an equivalence between KANs with spline basis functions and multichannel MLPs with power ReLU activations through a linear change of basis. We then analyze how this change of basis affects the geometry of gradient-based optimization with respect to spline knots. The KANs change-of-basis motivates a multilevel training approach, where we train a sequence of KANs naturally defined through a uniform refinement of spline knots with analytic geometric interpolation operators between models. The interpolation scheme enables a ``properly nested hierarchy'' of architectures, ensuring that interpolation to a fine model preserves the progress made on coarse models, while the compact support of spline basis functions ensures complementary optimization on subsequent levels. Numerical experiments demonstrate that our multilevel training approach can achieve orders of magnitude improvement in accuracy over conventional methods to train comparable KANs or MLPs, particularly for physics informed neural networks. Finally, this work demonstrates how principled design of neural networks can lead to exploitable structure, and in this case, multilevel algorithms that can dramatically improve training performance.

VISA: Value Injection via Shielded Adaptation for Personalized LLM Alignment

Aligning Large Language Models (LLMs) with nuanced human values remains a critical challenge, as existing methods like Reinforcement Learning from Human Feedback (RLHF) often handle only coarse-grained attributes. In practice, fine-tuning LLMs on task-specific datasets to optimize value alignment inevitably incurs an alignment tax: the model's pre-calibrated value system drifts significantly due to latent bias absorption from training data, while the fine-tuning process also causes severe hallucinations and semantic information loss in generated responses. To address this, we propose VISA (Value Injection via Shielded Adaptation), a closed-loop framework designed to navigate this trade-off. VISA's architecture features a high-precision value detector, a semantic-to-value translator, and a core value-rewriter. The value-rewriter is trained via Group Relative Policy Optimization (GRPO) with a composite reward function that simultaneously optimizes for fine-grained value precision, and the preservation of semantic integrity. By learning an optimal policy to balance these competing objectives, VISA effectively mitigates the alignment tax while staying loyal to the original knowledge. Our experiments demonstrate that this approach enables precise control over a model's value expression while maintaining its factual consistency and general capabilities, significantly outperforming both standard fine-tuning methods and prompting-based baselines, including GPT-4o.