Archive, newest first. Filter by topic, impact and company →
Problem This work addresses the limitations of sample efficiency in reinforcement learning (RL) for real-robot tasks, specifically in the context of juggling multiple balls. The authors highlight that traditional scalar…
Problem This work addresses the gap in the literature regarding the capability of fixed-size neural networks to achieve arbitrary accuracy in Sobolev approximation, specifically for functions in Sobolev spaces \(W^{s,\infty}((a,b)^d)\).…
Problem — The paper addresses the gap in understanding how documentation practices in AI research have evolved over the past decade, particularly in light of the reproducibility crisis. It highlights…
Problem This study addresses the gap in understanding the actual contribution of textual reviews in recommender systems, particularly when strong collaborative filtering baselines are employed. Despite the growing trend of…
Problem This work addresses the insufficient understanding of low frame rate degradation in neural audio codecs, particularly in the context of autoregressive speech synthesis. While prior research has shown that…
Problem The paper addresses the gap in predictive modeling of cryptocurrency implied-volatility surfaces, particularly for Bitcoin (BTC) and Ethereum (ETH). Existing methods, such as parametric smile re-fits, struggle with high…
Problem The paper addresses the limitations of existing multi-camera depth prediction methods in autonomous driving, particularly the challenges posed by low-overlap views from vehicle-mounted camera rigs. Traditional approaches rely on…
Problem The paper addresses the critical gap in auditing synthetic data for privacy disclosures, particularly in the context of generative AI and Large Language Models (LLMs). As synthetic data becomes…
Problem The paper addresses the limitations of traditional time-domain analysis in nanopore sensing, which is hindered by stochastic translocation dynamics that distort encoded information. The authors highlight a gap in…
Problem The paper addresses a significant gap in the literature regarding the situational engagement of Theory of Mind (ToM) in artificial intelligence systems, particularly in conflict scenarios. While existing AI-ToM…
Problem The paper addresses the challenge of mechanistic interpretability in large language models (LLMs), specifically the difficulty in interpreting learned circuits due to the polysemantic nature of raw neurons. Existing…
Problem The paper addresses the gap in real-time semantic mapping for autonomous rovers, particularly in the context of integrating perception and navigation under partial observability. Existing methods often lack the…
Problem This work addresses the underexplored behavioral properties that contribute to effective reasoning with Code Interpreters (CIs) in large language models (LLMs). Despite the increasing use of CIs for enhancing…
Problem Reinforcement learning (RL) systems often experience performance degradation when faced with distributional shifts, which can occur between training and evaluation phases or within non-stationary environments. Existing literature primarily addresses…
Problem The paper addresses the limitations of existing time-series forecasting models in true cold-start scenarios, where new items lack historical data. Traditional models rely on historical correlations, which are ineffective…
Problem Simulation-based inference (SBI) often suffers from misspecification, where discrepancies between simulated and real-world observations arise due to modeling simplifications. Existing methods, such as RoPE, require ground-truth parameter calibration pairs,…
Problem This paper addresses the significant variability in circuit discovery methods used for mechanistic interpretability of large language models (LLMs), specifically focusing on the state-of-the-art method EAP-IG. The authors identify…
Problem This work addresses a critical gap in the literature regarding the unintended consequences of visible reward proxies in reinforcement learning (RL) systems. The authors highlight that deployed agents often…
Problem — The paper addresses the lack of comprehensive datasets focused on social influence in adolescent communication, particularly in the context of interpersonal, media-based, and digital interactions. Existing datasets often…
Problem The paper addresses the inefficiencies in current sampling procedures for diffusion large language models (dLLMs), which typically rely on a fixed number of reverse denoising steps. This approach often…
Problem This work addresses the lack of a unified generative model for the natural sciences, which has traditionally relied on domain-specific architectures and methodologies. The authors highlight the need for…
Problem This work addresses the gap in open-source spatial question answering (SQA) systems for service robots navigating long egocentric routes. Existing methods predominantly rely on closed-source models like GPT-4o, which…
Problem The paper addresses the critical gap in the ability of vision-language models (VLMs) to appropriately refuse unanswerable queries in embodied agents, particularly in scenarios where overconfidence can lead to…
Problem This work addresses the gap in understanding how different large language model (LLM) architectures encode high-level concepts structurally. The authors highlight a geometric-functional universality dissociation, where moderate geometric convergence…
Problem This work addresses the limitations of traditional digital hardware in implementing neural dynamical systems (NDS), which are adept at modeling continuous-time dynamics but struggle with the discrete nature of…
Problem This work addresses the gap in converting formal mathematical proofs into human-readable text without sacrificing precision. Traditional proof systems have limited capabilities in this regard, often relying on syntactic…
Problem The paper addresses a significant gap in the literature regarding the evolving nature of federated learning (FL) communications. Traditional definitions primarily focus on the exchange of model weights and…
Problem This preprint addresses a critical gap in understanding the limitations of large language models (LLMs) in electronic health record (EHR) question answering. While aggregate accuracy metrics are often reported,…
Problem The paper addresses the gap in existing generalization bounds for deep learning models, particularly in safety-critical applications, where robustness and generalization are paramount. Current robustness-based generalization bounds often yield…
Problem This work addresses the lack of longitudinal analysis of online scam behaviors, specifically the temporal dynamics of scam types and their interrelations, which have not been comprehensively studied in…
