Research Papers

arXiv:2605.28612v1 Announce Type: new Abstract: Parity functions are fundamental Boolean operations with critical applications across machine learning, cryptography, and error correction. Yet, learning high-dimensional parity functions poses significant challenges: in a general setting, standard neural network architectures typically require exponential sample complexity, making gradient-based optimization intractable for large number of inputs $N$. We demonstrate that compact product-based neural architectures combined with stochastic data sparsity (Bernoulli inputs with $p_e \leq 1/N$) and appropriate hyperparameter choice enable efficient parity learning, with theoretical guarantees of convergence. Experiments validate our theory across dimensions up to $N = 100{,}000$, with empirical evidence showing optimal hyperparameter choices for $p_e$ and learning rate $\alpha$, as well as polynomial complexity scaling laws. This work establishes fundamental connections between architectural

arXiv:2605.27441v1 Announce Type: cross Abstract: Query understanding in large-scale industrial search systems is typically implemented as a cascade of disparate, task-specific components. While individually optimizable, this fragmented architecture incurs high maintenance overhead and results in inconsistent behaviors, particularly for long-tail queries. In this work, we propose and deploy a unified structured query understanding system that consolidates these heterogeneous functions into a single Small Language Model (SLM) that performs schema-constrained generation. To address the data bottlenecks inherent in unified modeling, we introduce Query Illuminator, a dual-purpose framework serving as: (i) a teacher model for high-quality auto-annotation and distillation, and (ii) a surrogate judge for scalable evaluation where human labels are scarce. We validate this approach through extensive offline and online tests within LinkedIn's Job Search system. Furthermore, we demonstrate the fr

arXiv:2605.28222v1 Announce Type: cross Abstract: We study quality-latency-resource trade-offs in a documentation-grounded retrieval-augmented generation (RAG) system that uses Low-Rank Adaptation (LoRA) of the generator. We build a manually verified benchmark of 5,144 question-answer pairs over the official Kubernetes documentation and combine it with a fixed hybrid-retrieval pipeline (BGE-M3 dense, BGE-M3 native sparse, Reciprocal Rank Fusion, cross-encoder reranking). Over this benchmark we ablate 20 LoRA configurations on Llama-3.2-3B-Instruct and Llama-3.1-8B-Instruct across rank and target-module choices, and evaluate each on token-level F1, LLM-judged groundedness and correctness (pass@4), inference latency, inference memory, and training cost, all reported with bootstrap 95% confidence intervals. Pareto analysis shows that LoRA adapters acting only on the q and v attention projections consistently dominate the front, while the 3B/8B choice mainly defines operating regime. A par

arXiv:2510.24941v4 Announce Type: replace Abstract: Large language models can generate long chain-of-thought (CoT) reasoning, yet prior work suggests that CoT can be post-hoc rationalization rather than a faithful reflection of the computation through explicitly designed settings. In this work, we go further and propose a True Thinking Score (TTS) to quantify the causal contribution of each step in CoT to the model's final prediction in realistic reasoning problems. Across eleven models ranging from 1.5B to 1.1T parameters on common reasoning benchmarks, we find that CoTs often interleave true-thinking steps, which causally affect the final answer, with decorative-thinking steps, which appear useful but have little causal influence; Such decorative steps remain prevalent even for frontier models: Over 30% of steps in Kimi-K2.6 are decorative on MATH with TTS <= 0.005. Furthermore, TTS enables effective CoT pruning: removing 50% of CoT steps with the lowest TTS can largely maintain the

arXiv:2602.02417v2 Announce Type: replace Abstract: Continual learning aims to acquire tasks sequentially without catastrophic forgetting, yet standard strategies face a core tradeoff: regularization-based methods (e.g., EWC) can overconstrain updates when task optima are weakly overlapping, while replay-based methods can retain performance but drift due to imperfect replay. We study a hybrid perspective: \emph{trust region continual learning} that combines generative replay with a Fisher-metric trust region constraint. We show that, under local approximations, the resulting update admits a MAML-style interpretation with a single implicit inner step: replay supplies an old-task gradient signal (query-like), while the Fisher-weighted penalty provides an efficient offline curvature shaping (support-like). This yields an emergent meta-learning property in continual learning: the model becomes an initialization that rapidly \emph{re-converges} to prior task optima after each task transitio

arXiv:2605.01046v3 Announce Type: replace Abstract: LoRA adapts large language models (LLMs) by restricting updates to low-rank subspaces of pre-trained weights. While this substantially reduces training cost, the effectiveness of adaptation critically depends on which subspace is chosen at initialization: a poor initialization that allocates capacity to task-irrelevant directions can severely hinder downstream performance. Existing initialization strategies primarily rely on the intrinsic properties of pre-trained weights, implicitly assuming that weight geometry alone reflects task relevance. However, such criteria overlook how the model interacts with the downstream data distribution. In this work, we formulate LoRA initialization as identifying the degree of impact of directions in parameter space under the target data distribution. We argue that data-aware sensitivity, rather than weight-only magnitude, should govern the choice of adaptation subspaces. Building on this perspective

arXiv:2605.08678v2 Announce Type: replace Abstract: Modern AI progress has been driven by ML methods that are generalizable across settings and scalable to larger regimes. As large language models demonstrate advanced capabilities in reasoning, coding, and engineering tasks, it is increasingly important to understand whether they can discover such methods rather than only apply existing ones. We introduce MLS-Bench, a benchmark for evaluating whether AI systems can invent generalizable and scalable ML methods. MLS-Bench contains 140 tasks across 12 domains, each requiring an agent to improve one targeted component of an ML system or algorithm and demonstrate that the improvement generalizes across controlled settings and scales. We find that current agents remain far from reliably surpassing human-designed methods, and that engineering-style tuning is easier for them than genuine method invention. We further study the effects of test-time scaling, adaptive compute allocation, and conte

arXiv:2510.06970v2 Announce Type: replace-cross Abstract: Compliance with maritime traffic rules is essential for the safe operation of autonomous vessels, yet training reinforcement learning (RL) agents to adhere to them is challenging. The behavior of RL agents is shaped by the training scenarios they encounter, but creating scenarios that capture the complexity of maritime navigation is non-trivial, and real-world data alone is insufficient. To address this, we propose a falsification-driven RL approach that generates adversarial training scenarios in which the vessel under test violates maritime traffic rules, which are expressed as signal temporal logic specifications. Our experiments on open-sea navigation with two vessels demonstrate that the proposed approach provides more relevant training scenarios and achieves more consistent rule compliance.

arXiv:2510.22190v2 Announce Type: replace-cross Abstract: Bent radio active galactic nuclei (RAGNs) -- wide-angle tails (WATs) and narrow-angle tails (NATs) -- trace dense environments in galaxy groups and clusters, yet no multiclass classifier simultaneously separates them from straight Fanaroff--Riley types (sFRI, sFRII) using visually inspected labels and unlabelled data. We release FIRST-2060, a four-class labelled dataset of 2060 RAGNs (sFRI, sFRII, WAT, NAT) constructed from three publicly available catalogues through multi-tier visual inspection, together with the semi-supervised RGC 1.0 model that leverages 20,000 unlabelled sources. We benchmark RGC against five supervised baselines. FIRST-2060 is provided in two preprocessing variants: $\mathbf{R}_{L1}$, which retains spurious sources, and $\mathbf{R}_{L2}$, from which they are removed. The RGC model integrates the self-supervised framework BYOL (Bootstrap Your Own Latent) with an $E(2)$-equivariant steerable CNN (E2CNN) enco

arXiv:2605.27386v1 Announce Type: new Abstract: Advancements in augmented reality (AR) technologies offer immense potential for mobile experiences. However, most commercial and educational AR systems assume a baseline of predictable user behavior and stationary interaction. Preschoolers and children in early childhood education, specifically ages 3 to 8, are naturally erratic, physically dynamic, and prone to rapid locomotion, making them the ultimate stress test for mobile spatial computing. Through a focused analysis of recent literature on physical activity and spatial learning in AR for preschoolers, this paper identifies points of friction in current mobile deployments. We highlight recurring failures in camera tracking during dynamic movement, physical safety hazards caused by screen-induced distraction, spatial crowding around physical markers, and the privacy risks of continuous environmental surveillance. To address these challenges, we propose AnchorPlay AR, a conceptual prot

arXiv:2605.27554v1 Announce Type: new Abstract: Infographic designers balance many choices at once: chart type, color, and whether to add a benchmark or a scale. Past work studies these factors one at a time, so we know little about how readers weigh them against each other. We address this gap with a choice-based conjoint study (N = 65) in which participants viewed pairs of infographics on a mock newspaper page about unemployment. Each infographic varied across three attributes: comparison type (none, US average, percentage scale), color (red, blue), and graphic type (single icon, icon series, bar chart). Comparison type drove most of the preference variation (58.5%), followed by graphic type (29.2%) and color (12.3%). Readers favored percentage scale markers and benchmark comparisons; color had no practical effect. The percentage scale level adds axis information rather than a benchmark, so the comparison type result mixes two distinct ideas. A single topic and a narrow palette also

arXiv:2605.28571v1 Announce Type: new Abstract: Despite warnings that LLMs can make mistakes, users often develop inappropriate trust and accept incorrect answers without critical evaluation. Uncertainty quantification (UQ), displaying LLMs' confidence, has emerged as a promising approach to calibrate user trust. However, prior empirical studies on uncertainty communication have treated uncertainty as a single numerical score or simple natural language expression. This simplification fails to capture a key property of LLM outputs: a single response often comprises multiple claims and reasoning steps, each with distinct levels of uncertainty. To address this gap, this study investigates uncertainty granularity (i.e., the extent to which uncertainty is expressed at different levels within an LLM response) and examines its impact on LLM-assisted decision-making. We conducted a large-scale, between-subjects study (N=192) in which participants answered medical questions using LLMs that disp

arXiv:2605.27864v3 Announce Type: replace Abstract: Large language models (LLMs) are increasingly applied in finance, yet most existing work emphasizes trading signals or financial NLP tasks centered on prediction. Institutional fundamental research, by contrast, requires human analysts or AI agents to gather evidence, identify business drivers, compare competing viewpoints, and generate investment memos. Its broader goal is not merely to predict outcomes, but to produce investment plans that are transparent, reusable, and verifiable, while contributing to the cumulative development of investment knowledge. We present FundaPod, a multi-persona agent platform for AI-assisted fundamental investment research. We argue that fundamental research is a human-centric decision-support task that is qualitatively distinct from trading-signal generation, and is therefore better served by an independence-preserving architecture. In FundaPod, AI agents with different personas, such as value investor

arXiv:2605.27458v2 Announce Type: replace-cross Abstract: Transformer has significantly propelled the development of artificial intelligence, and certainly the development of agents as well. We categorize attention structures of Transformer into two types based on the source of the input information: homogenous and heterogenous attention structures. Heterogenous attention structures, with co-attention as a typical example, process information from different sources. Heterogenous attention structure is the foundation for Transformer models to achieve more complex functions and integrate more modal information. Whether for research purposes or policy requirements, the interpretation of Transformer models with heterogenous attention structures is an important task. The fusion of information from different sources brings new challenges. Our work mainly includes two parts: method and experimentation. In terms of method, we propose an interpretation method for Transformer models with heterog

arXiv:2605.28119v2 Announce Type: replace Abstract: Colo-segment recognition in colonoscopy videos is a key requirement for many downstream tasks, but existing automatic recognition methods only use colonoscopy images without fully exploiting the use of temporal information, leading to poor performance. Additionally, relevant public video-based datasets are in scarcity. To tackle this problem, we curate and release a labeled dataset specifically for the task of colo-segment recognition. In addition, we propose a two-stage deep learning-based framework, Colo-Segment Recognition via SpatioTemporal Network (ST-ColoNet), for the task of colo-segment recognition from colonoscopy videos which includes the Colorlaus module that uses metric learning to optimize edge-mediated spatial feature extraction, as well as the Full-Temp module which combines three self-attention patterns to better approximate full self-attention on long colonoscopy sequences and optimize temporal feature aggregation. Th