Archive, newest first. Filter by topic, impact and company →
Problem This work addresses the gap in understanding the fitness progress of Evolution Strategies (ES) in generic optimization problems, particularly when the fitness landscape is intractable to model directly. The…
Problem The paper addresses the latency issues inherent in traditional audio commentary systems for live gameplay, which typically operate in a sequential manner—capturing frames, generating text, and synthesizing speech one…
Problem Current skill self-evolution methods for large language model (LLM) agents are limited by their reliance on single execution trajectories per task, which restricts the diversity of learned skills. Existing…
Problem The paper addresses the limitations of existing Reinforcement Learning with Verifiable Rewards (RLVR) methods in Large Language Models (LLMs), particularly their tendency to produce unnecessarily long reasoning rollouts that…
Problem Existing research on self-supervised foundation models in MRI has predominantly focused on segmentation and dense prediction tasks, leaving a gap in systematic investigations for disease detection. This paper addresses…
Problem The paper addresses the challenge of generating and modeling facial micro-expressions, which are critical for understanding genuine human emotions but are difficult to capture due to their subtlety and…
Problem The paper addresses the challenge of contextual anomaly detection in maritime environments, particularly under conditions of highly imbalanced context distributions. Existing models often struggle with rare context regimes, leading…
Problem — This work addresses the inadequacy of traditional Turing Tests in evaluating AI systems, particularly in terms of trustworthiness rather than mere indistinguishability. The authors propose a new interactive…
Problem The paper addresses the limitations in existing speech-driven talking character animation methods, which struggle to balance accurate lip synchronization with dynamic facial expressions and head movements. Prior approaches often…
Problem This work addresses the unexplained prevalence of specific narrative themes in large language models (LLMs), particularly the recurring character of 'Elias Thorne,' a lighthouse keeper. The authors highlight a…
Problem Recent advancements in diffusion-based image editing have focused on instruction-driven modifications while maintaining a strong conditioning on the source image throughout the denoising process. However, this persistent conditioning can…
Problem The paper addresses the limitations of traditional coastal wave monitoring systems, which are often expensive and have restricted spatial coverage. Existing deep learning methods for passive ocean monitoring lack…
Problem This work addresses the limitations of existing post-training quantization (PTQ) methods, which often lack a systematic approach to sensitivity analysis. Current techniques typically couple sensitivity analysis with quantization procedures,…
Problem The paper addresses a significant gap in the literature regarding the causal effects of agentic AI coding tools on software architecture, particularly in the context of "vibe coding." While…
Problem The paper addresses the limitations of traditional classification trees in stratified contexts, particularly when controlling for confounding variables such as temporal, spatial, or demographic factors. Existing tree growth procedures…
Problem This paper addresses the lack of unified multimodal models (UMMs) that effectively integrate image and video tokenization within a single architecture. Existing models often treat image and video modalities…
Problem This work addresses the limitations of existing contrastively trained vision-language models (VLMs) like CLIP, which struggle with compositional understanding due to their "bag-of-words" behavior. Specifically, these models fail to…
Problem The paper addresses the limitations of asynchronous stochastic gradient descent (ASGD) in distributed and federated learning settings, particularly the negative impact of stragglers—slow workers that delay updates. While ASGD…
Problem The paper addresses the challenge of simultaneous forecasting in multi-system environments, where traditional methods typically predict one system at a time. This limitation can lead to inefficiencies and inaccuracies…
Problem This paper addresses the gap in explainability within self-supervised classification-based time series anomaly detection (TSAD) methods. While these approaches have demonstrated strong performance by leveraging transformation-specific patterns, they often…
Problem This work addresses the gap in understanding how different transformer modules may benefit from distinct weight-space geometries during optimization. The authors highlight that existing literature typically applies uniform manifold…
Problem This work addresses the challenges of deploying an AI-powered mobile guide in a museum setting, specifically for the Grand Egyptian Museum (GEM). The authors identify three key issues: (1)…
Problem The paper addresses the challenge of identifying latent dynamical systems from high-dimensional, noisy measurements, a critical issue in representation learning, system identification, and scientific discovery. The authors highlight a…
Problem The paper addresses the challenge of operationalizing pluralism in large language models (LLMs), particularly the need to identify and represent diverse perspectives in generated text. Despite the growing interest…
Problem The paper addresses the challenge of automatically detecting mounds on Mars from Digital Elevation Models (DEMs), a task traditionally performed through manual mapping of morphological parameters. This gap in…
Problem This work addresses the challenge of optimizing large-scale AI models through efficient quantization techniques, specifically for Mixture of Experts (MoE) architectures. The authors highlight a gap in existing literature…
Problem This work addresses the limitations of existing control mechanisms in diffusion models, particularly the reliance on implicit target distributions defined through sampling rules or heuristic energy functions. The authors…
Problem Existing computer-use agents face limitations in professional software manipulation, particularly in GUI-based and API-based approaches. GUI agents suffer from fragile visual grounding and long-horizon error accumulation, while API-based methods…
Problem Existing automated tools for passive acoustic monitoring in ecology are often narrowly trained and lack transferability across species. This paper addresses these limitations by proposing a semi-supervised, multi-task framework…
Problem This work addresses the limitations of applying NVFP4 (NVIDIA's Variable-Precision Floating Point 4) quantization to large reasoning models (LRMs), which suffer from degraded reasoning accuracy and suboptimal latency during…
