Archive, newest first. Filter by topic, impact and company →
Problem The paper addresses the inefficiency of uniform token processing in time series (TS) language models, which arises from the fundamentally different information structures of TS tokens and prompt tokens.…
Problem — This work addresses the limitations of traditional shielded reinforcement learning, which is often viewed as a runtime safety mechanism. The authors argue that the automata-theoretic processes involved in…
Problem — This work addresses the gap in understanding the optimality of voting schemes in the context of the Probably Approximately Correct (PAC) learning framework. Specifically, it focuses on the…
Problem This work addresses the emerging risk of generative recommenders inadvertently promoting fake products due to polluted web content retrieved during the recommendation process. As search-augmented large language models (LLMs)…
Problem The rapid advancement of agent systems across various domains has led to fragmented evaluation methodologies, primarily reliant on fixed, LLM-centric benchmarks that necessitate extensive integration and often result in…
Problem This preprint addresses the gap in understanding the reasoning capabilities of large language models (LLMs) compared to human reasoning. It challenges the prevailing notion that LLMs fail due to…
Problem This work addresses the gap in robust trajectory optimization under uncertainty, specifically focusing on scenarios where the uncertainty is not strictly Gaussian. The authors highlight the limitations of existing…
Problem The paper addresses the challenge of adapting dispatch objective weights in three-sided marketplaces, specifically in the context of food delivery services like DoorDash. Traditional reinforcement learning approaches often struggle…
Problem — The paper addresses the gap in understanding the causal influence of individual steps in chain-of-thought (CoT) reasoning within large language models. Despite CoT being a prevalent method for…
Problem This work addresses the lack of verifiable benchmarks for evaluating AI agents in the context of epigenomics analysis, specifically focusing on short-horizon decision-making tasks. The authors highlight that existing…
Problem The paper addresses the challenge of training orchestrators for Multi-Agent Systems (MAS) built on Large Language Models (LLMs), which is often limited by the need for extensive human supervision…
Problem The paper addresses a significant gap in the literature regarding the aggregation of confidence signals in multiagent systems, particularly in Natural Language Processing (NLP). While prior work has explored…
Problem The paper addresses the limitations of traditional bagging ensembles, such as Random Forests and Bagged Neural Networks, which utilize uniform voting power across base estimators. This approach neglects the…
Problem The paper addresses the limitations of existing deep learning approaches for automated waste recycling (AWR), which often depend on large backbone networks that are computationally inefficient and exhibit performance…
Problem — This paper addresses the gap in understanding how mental health content is framed by creators and perceived by audiences on TikTok, particularly during Mental Health Awareness Month. Despite…
Problem This work addresses the gap in video super-resolution (VSR) techniques that primarily focus on motion refinement, neglecting the enhancement of texture quality. While previous methods have utilized event-based vision…
Problem The paper addresses the gap in the capability of AI systems to execute scientific protocols in laboratory environments, where existing Vision-Language-Action (VLA) models have primarily been trained on household…
Problem This work addresses a significant gap in learning theory regarding the generalization capabilities of algorithms when faced with strongly dependent data. Most existing results are predicated on the assumption…
Problem The paper identifies a significant gap in the capabilities of existing Multimodal Large Language Models (MLLMs) in the healthcare domain, particularly for multilingual and low-resource contexts. Current models predominantly…
Problem — The paper addresses the gap in time series forecasting methods that assume future observation timestamps are known, which is often unrealistic in real-world scenarios. Existing techniques, including Neural…
Problem — The paper addresses the lack of gauge-invariant neural network architectures capable of modeling observables in physics and geometry, specifically those represented as cup products of cochains. The authors…
Problem Current literature on AI in scientific discovery predominantly emphasizes two capabilities: knowledge search and execution via optimization and automation. However, these aspects overlook the essential process of model formation,…
Problem The paper addresses the gap in reward-guided fine-tuning for any-length discrete diffusion models, a topic that remains largely unexplored in the literature. Existing methods primarily focus on fixed-length sequences,…
Problem This work addresses the challenge of generalizing deep learning models for magnetic resonance (MR) reconstruction from adult to neonatal populations, a gap in the literature that remains largely unaddressed.…
Problem The paper addresses the challenge of efficiently implementing Spiking Neural Networks (SNNs) in hardware, which is hindered by high power and area costs associated with neuronal computations. Additionally, the…
Problem The paper addresses the limitations of existing text-guided image editing methods using visual autoregressive (VAR) generators, which primarily manipulate token streams or flat logits. These methods do not leverage…
Problem Personalized health AI systems encounter a significant cold-start problem, where machine learning models require extensive individual behavioral data to differentiate between constitutional variations and environmentally driven deviations. This paper…
Problem The paper addresses the limitations of existing Retrieval-Augmented Generation (RAG) systems, particularly in handling long documents. Traditional RAG approaches struggle with the trade-off between context preservation and the introduction…
Problem Turn-taking in multi-party spoken interactions poses significant challenges for voice agents, particularly in dynamic environments with competing speakers and varying user expectations. Existing systems often struggle with accurately managing…
Problem Existing root cause analysis (RCA) techniques often rely on static rules, correlation heuristics, or topology-local reasoning, which are inadequate in dynamic environments where faults propagate through complex physical and…
