Archive, newest first. Filter by topic, impact and company →
Problem The paper addresses the challenge of accurately modeling chaotic, turbulent, and stochastic systems using ensemble predictions. Existing methods often rely on complex drift and diffusion estimations, which can be…
Problem Existing deep learning-based PET image denoising methods typically assume a fixed dose reduction factor (DRF), leading to performance degradation when faced with variable DRFs in practical applications. Preliminary studies…
Problem This work addresses the gap in the literature regarding the agnostic learning of general ReLU (Rectified Linear Unit) functions under the Gaussian distribution, specifically in the interactive learning setting.…
Problem The paper addresses a significant gap in the evaluation of video-based world models, particularly in assessing their physical faithfulness, geometric consistency, and interaction fidelity. Existing benchmarks primarily focus on…
Problem The paper addresses a gap in the literature regarding the effectiveness of synthetic post-training data curation methods, specifically the lack of examination of provenance grounding in filtering signals and…
Problem This work addresses the limitations of existing PPG-based blood pressure (BP) estimation models, which often rely on amplitude-dominated shortcuts and do not effectively incorporate demographic covariates that influence vascular…
Problem The paper addresses the limitations of Feedback Alignment (FA) in deep neural networks, particularly its inability to scale effectively compared to Backpropagation (BP). FA, which uses fixed random feedback…
Problem This work addresses the gap in football (soccer) pass evaluation methodologies by proposing a Monte Carlo Tree Search (MCTS)-like framework. Existing literature lacks a unified approach that integrates value…
Problem The paper addresses the limitations of existing reinforcement learning with verifiable rewards (RLVR) approaches, particularly in multi-turn scenarios where reward contrast is insufficient. Traditional methods focus on allocating rollout…
Problem This work addresses a significant gap in the literature regarding dynamic assortment problems on two-sided service platforms, particularly in scenarios where both customer and seller choice parameters are unknown.…
Problem The paper addresses the challenge of designing FPGA-based accelerators for AI workloads, which traditionally requires extensive domain knowledge and manual effort due to the complex design space involving architectural…
Problem This paper addresses the gap in understanding how generative AI disclosures in journalism affect reader trust. Current practices, which include either minimal one-line labels or extensive disclosures, fail to…
Problem Current deep learning approaches for brain tumor classification predominantly utilize single-modality data, typically MRI or CT images, which limits their ability to emulate the comprehensive diagnostic process employed by…
Problem This paper addresses the critical shortage of trained sonographers in low- and middle-income countries, where over half of pregnant women lack access to skilled ultrasound services. Current deep learning…
Problem The paper addresses the critical issue of hallucinations in language models (LMs), where models generate factually incorrect information. This phenomenon poses significant risks, particularly in high-stakes applications where reliance…
Problem This work addresses the gap in understanding the limitations of learning $\tanh$ neural networks when subjected to finite-precision computations. Previous research has primarily focused on ReLU networks, leaving a…
Problem This work addresses the gap in the literature regarding the effectiveness of Transformer architectures in network intrusion detection, particularly under realistic conditions. Previous studies often report high performance on…
Problem This work addresses the limitations of existing representation autoencoders (RAEs) built on pretrained vision foundation models (VFMs), particularly their suboptimal reconstruction quality due to insufficient preservation of fine-grained visual…
Problem The paper addresses the limitations of existing reinforcement learning (RL) approaches in humanoid soccer shooting, particularly the challenges of achieving stability and adaptability in whole-body movements. Traditional motion tracking-driven…
Problem This work addresses the challenge of integrating expressive continuous control policies, such as diffusion and flow models, into reinforcement learning (RL) frameworks. While these models excel in supervised imitation…
Problem This study addresses the gap in understanding how large language models (LLMs) exhibit cross-lingual distributional skew, termed the Shibboleth Effect, particularly under adversarial conditions. The authors highlight that existing…
Problem The paper addresses the limitations of existing decentralized pre-training methods for large language models (LLMs), particularly those that rely on synchronous All-Reduce operations. These methods can become bottlenecks in…
Problem The paper addresses significant gaps in the evaluation of interactive agents, particularly in the context of user simulations. Existing frameworks often rely on static benchmarks that inadequately capture the…
Problem This work addresses the limitations of existing critic models for Computer Use Agents (CUAs) that primarily focus on short-term decision-making and lack visual grounding capabilities. The authors highlight that…
Problem This work addresses the limitations of existing reward backpropagation methods in text-to-image flow matching models, particularly the inefficiencies caused by storing activations across the full sampling trajectory and the…
Problem This work addresses the gap in understanding and controlling the behavior of Multimodal Large Language Models (MLLMs) under complex personality conditions. Existing literature lacks a systematic evaluation framework for…
Problem Existing benchmarks for evaluating agentic systems, particularly those leveraging large language models (LLMs), are limited in their ability to assess multi-step reasoning and coordination across diverse domains. Current frameworks…
Problem Current neural population activity models are constrained by fixed read-in and readout layers tied to specific recorded neurons, limiting their adaptability in long-term brain-computer interfaces (BCIs). This paper addresses…
Problem The paper addresses the gap in understanding how frontier large language models (LLMs) respond to control interventions, particularly their ability to recognize when their outputs are being modified by…
Problem The paper addresses the challenge of predicting protein properties, such as binding affinity and thermostability, from sparse experimental data. This is a significant gap in the literature, particularly in…
