Archive, newest first. Filter by topic, impact and company →
Problem The paper addresses the gap in existing literature regarding the simulation of multi-turn LLM agent serving, which interleaves model calls with external tool invocations. Current LLM serving simulators primarily…
Problem The paper addresses the lack of automated Individualized Education Program (IEP) generation in Traditional Chinese, a domain that has been largely unexplored due to data scarcity, stringent privacy regulations,…
Problem The paper addresses a significant gap in the evaluation of privacy risks associated with medical language models (LMs), particularly in the context of their ability to memorize and disclose…
Problem The paper addresses the under-explored role of table representation in evaluating Large Language Models (LLMs) and Vision-Language Models (VLMs) on table reasoning tasks. Existing evaluations conflate content, format, layout,…
Problem The paper addresses the gap in reliable uncertainty estimation (UE) for code generation using large language models (LLMs). Existing UE methods primarily adapted from natural language (NL) generation overlook…
Problem This paper addresses the gap in evaluating user experience (UX) in AI assistants, which has been largely overlooked in favor of general model performance metrics. The authors highlight the…
Problem The paper addresses the significant gap in text-to-speech (TTS) capabilities for low-resource languages, which are often overlooked in favor of high-resource languages. Existing TTS models predominantly focus on a…
Problem This work addresses the gap in stylometric analysis methods that lack interpretability and robustness in author attribution. Existing techniques often fail to provide clear insights into the linguistic features…
Problem This work addresses the significant performance gap in automatic speech recognition (ASR) for Dravidian languages when using multilingual models like Whisper, which are optimized for high-resource languages. The authors…
Problem This work addresses the gap in the literature regarding the simultaneous analysis of mobility patterns and social media discourse during crises, which are often studied in isolation. The authors…
Problem The Traveling Salesman Problem (TSP) is a well-known NP-hard combinatorial optimization challenge that seeks the shortest Hamiltonian cycle visiting each city exactly once. This paper addresses the limitations of…
Problem This work addresses the limitations of local search heuristics in solving the vertex coloring problem, which is known to be NP-hard. The paper identifies specific structures in bipartite graphs…
Problem — This work addresses the gap in understanding emergent phenomena in multi-agent systems, particularly in economic simulations. The authors investigate how control mechanisms can influence emergent behaviors in a…
Problem The paper addresses a significant gap in the understanding of stopping criteria within the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), a leading black-box optimization algorithm. While CMA-ES employs multiple…
Problem This work addresses the lack of efficient optimization frameworks for SRAM design that simultaneously consider both architectural parameters and transistor sizing. Existing methods often optimize these aspects in isolation,…
Problem — This paper addresses the inadequacy of small language models (LMs) in performing rare tasks, which is attributed to the phenomenon where frequent tasks dominate the learning process, leading…
Problem This work addresses the gap in the literature regarding the fine-tuning of large language models (LLMs) for specific stylistic outputs, particularly in the context of technical writing from the…
Problem — This work addresses the challenge of catastrophic forgetting in continual learning, particularly in the context of parameter-efficient finetuning methods. Existing approaches often lead to interference with dominant singular…
Problem The paper addresses the gap in humanoid robot control systems that struggle with the synthesis of dense kinematic or spatial references from task semantics. Existing whole-body controllers often require…
Problem Existing methods for integrating repository-level context into code language models, such as retrieval-augmented generation (RAG) and per-repository fine-tuning, are inefficient and brittle, particularly in the face of evolving codebases.…
Problem This paper addresses the limitation of existing Vision-Language-Action (VLA) models, which typically operate at a fixed execution speed derived from training demonstrations. Prior methods for enhancing speed, such as…
Problem The paper addresses the inadequacy of traditional external regret metrics in online learning when applied to repeated games with adaptive opponents. Existing regret definitions do not account for the…
Problem Existing 3D multimodal large language models (3D-MLLMs) primarily focus on object-centric representations, which limits their effectiveness in understanding fine-grained part structures necessary for nuanced interactions within 3D environments. This…
Problem The paper addresses a significant gap in the literature regarding the detection of AI-generated text, particularly in the context of progressive human-AI co-editing. Existing benchmarks primarily focus on final…
Problem The paper addresses the challenge of training agents in multi-turn simultaneous bidding scenarios within partially observable n-player games, a gap in the literature concerning equilibrium strategies in competitive environments.…
Problem The paper addresses the limitations of standard backpropagation through time (BPTT) in training recurrent neural networks (RNNs), particularly its sequential nature, which restricts parallelism, and its susceptibility to vanishing…
Problem — This work addresses the inefficiency of monolithic architectures in continuous-time generative models, which struggle to manage diverse signal regimes effectively. The authors highlight the limitations of existing methods…
Problem This work addresses the limitations of Vision-Language Models (VLMs) in spatial reasoning, particularly their inability to infer unobserved layouts and maintain cross-view consistency when only limited egocentric observations are…
Problem The paper addresses the limitations of existing reinforcement learning (RL) fine-tuning methods for reasoning language models, particularly those relying on Group Relative Policy Optimization (GRPO) and its variants. These…
Problem This work addresses the limitations of discrete diffusion language models in retrieval-augmented generation (RAG) by leveraging discarded low-confidence tokens during the denoising process. The authors identify a gap in…
