Archive, newest first. Filter by topic, impact and company →
Problem This work addresses the limitations of existing Koopman operator learning methods, particularly the trade-off between expressiveness and structure preservation in finite-dimensional observable choices. Current approaches either rely on fixed…
Problem The paper addresses the limitations of existing supervised sequence models in individual-level mobility prediction, which require task-specific training and lack decision-level transparency. While recent LLM-based methods have improved interpretability,…
Problem Current Large Audio Language Models (LALMs) operate in an offline manner, limiting their interactivity and responsiveness. Existing streaming audio models are typically designed for single tasks, such as automatic…
Problem This work addresses the gap in continual learning frameworks that effectively model how children learn word meanings from egocentric experiences. Prior approaches, while demonstrating the ability to learn from…
Problem The paper addresses the challenge of adapting vision foundation models to specialized scientific domains where labeled data is scarce. Traditional supervised fine-tuning often leads to a loss of generality…
Problem This work addresses the gap in the literature regarding the application of human pedagogical methods to improve arithmetic reasoning in language models. Specifically, it investigates whether structured teaching techniques…
Problem The paper addresses significant gaps in the evaluation of large language models (LLMs) by introducing KINA, a benchmark designed to operationalize disciplinary representativeness, which has been inadequately addressed in…
Problem The challenge of generating concise and informative titles for research papers is a persistent issue in academic writing, often leading to suboptimal title selection by authors. This paper addresses…
Problem Current benchmarks for frontier models predominantly assess single-turn responses or short-horizon tasks, neglecting the complexities of long-horizon iterative processes essential in scientific and engineering domains. This paper addresses this…
Problem Current research in Computer-Aided Design (CAD) often focuses on isolated tasks, leading to a fragmented understanding of multi-modal and multi-task learning capabilities. There is a notable absence of a…
Problem The paper addresses a significant limitation in the self-consistency method for large language models (LLMs), which relies on majority voting to select answers from multiple sampled reasoning paths. This…
Problem This work addresses the limitations of existing Graph Neural Networks (GNNs) and Graph Transformers (GTs) in capturing intermediate-scale graph structures. Current models primarily focus on local edges or global…
Problem This work addresses the gap in the literature regarding the identification of control-affine reduced-order models (ROMs) using high-dimensional state and input data. Existing methods often struggle with the complexity…
Problem This work addresses the gap in the literature regarding the practical application of declarative interaction protocols in multiagent systems, particularly in the context of Agentic AI. The authors highlight…
Problem The paper addresses the limitations of existing marginal inference methods in discrete graphical models, particularly the trade-off between exactness and scalability. Exact algorithms become intractable for high-treewidth graphs, while…
Problem This work addresses the gap in AI agent recovery mechanisms when encountering API validation errors. Traditional APIs provide verbose error messages that lack actionable guidance for agents, necessitating external…
Problem The paper addresses the limitations of existing streaming 3D reconstruction models, which typically rely on a fixed coordinate system tied to the initial frame or a persistent scene memory.…
Problem Generative visual models have historically struggled with precise spatial control, particularly in mapping numerical coordinates to 2D image canvases. This limitation hinders the ability to generate images with specific…
Problem The paper addresses the limitations of autoregressive chain-of-thought (CoT) reasoning in large language models (LLMs), which is inherently unidirectional. This unidirectional inductive bias leads to error snowballing, where an…
Problem The paper addresses the growing reliance on non-ideal experimental strategies in foundation model research due to the prohibitive costs of controlled experiments. It highlights the lack of comprehensive frameworks…
Problem Large language models (LLMs) exhibit a significant vulnerability to shortcut learning, particularly when faced with out-of-distribution (OOD) inputs that differ semantically from their training data, despite sharing identical logical…
Problem This work addresses the limitations of the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm in multi-agent reinforcement learning (MARL) settings, particularly the challenges of non-stationarity and inter-agent coordination. The…
Problem The validation of handwritten signature lists for Swiss popular initiatives is a labor-intensive manual process, which poses a significant bottleneck in democratic participation. This paper addresses the gap in…
Problem The paper addresses the challenge of merging task-specific and domain-specific LoRA (Low-Rank Adaptation) adapters into a unified model, a largely unexplored area in the literature. Existing approaches treat these…
Problem The paper addresses the limitation of traditional Transformer architectures, where information flow across layers is restricted to simple residual connections. This results in later layers being unable to selectively…
Problem Existing approaches to cross-view geo-localization either focus on large-scale image retrieval or precise pose estimation, but not both. Retrieval methods allow for wide-area searches but sacrifice localization accuracy, while…
Problem The paper addresses the limitations of large language models (LLMs) in deontic reasoning, particularly their struggles with long, cross-referenced rulesets that are essential for accurate reasoning in legal and…
Problem The paper addresses a significant gap in the evaluation of memory capabilities in multi-modal models, particularly in the context of long-form video understanding. While existing benchmarks focus on perception…
Problem The paper addresses the critical issue of user prompt privacy in the context of public large language models (LLMs) like ChatGPT. Existing privacy-preserving methods often compromise either utility or…
Problem The paper addresses the limitations of existing multi-agent reinforcement learning (MARL) frameworks, which often rely on task-specific reward designs that lack a robust grounding in the interaction structures among…
