13603 papers
This paper introduces a post-hoc framework to explain, verify, and align the semantic hierarchies in vision-language model embeddings, revealing that while image encoders offer superior discriminative power, text encoders better align with human taxonomies, highlighting a trade-off between zero-shot accuracy and ontological plausibility.
The paper proposes the Dual-Scale Neural Operator (DSO), a novel architecture that decouples local feature extraction and global trend aggregation to effectively address the long-term stability and precision challenges in fluid dynamics forecasting, achieving state-of-the-art results with over 88% error reduction compared to existing methods.
The paper proposes L0GM, a modality-agnostic framework that employs hard-concrete stochastic gating to enforce L0-style sparsity on learned representations across graphs, language, and tabular data, thereby enabling unified, efficient, and reliable cross-modality learning with improved probability calibration.
This paper introduces ViPAC, a dual-branch learning framework that generates natural language descriptions from vibrotactile signals by disentangling their periodic and aperiodic components, and validates the approach using the newly constructed LMT108-CAP dataset.
Inspired by cognitive science's view of problem-solving as a search over structured spaces, GroupRAG introduces a framework that identifies latent problem groups to enable fine-grained, multi-perspective retrieval and reasoning, thereby outperforming traditional RAG and Chain-of-Thought baselines on MedQA.
This paper establishes a reproducible benchmark for implicit neural representations on the MapZebrain larval zebrafish atlas, demonstrating that explicit spectral encodings like Haar and Fourier features outperform smoother alternatives in preserving high-frequency neuroanatomical boundaries and fine neuronal processes.
This paper proposes Hybrid Document-Routed Retrieval (HDRR), a two-stage architecture that resolves the robustness-precision trade-off in financial RAG systems by combining Semantic File Routing for document filtering with chunk-based retrieval, thereby achieving superior performance in accuracy, failure reduction, and perfect-answer rates on the FinDER benchmark.
The paper introduces PiCSRL, a physics-informed contextual spectral reinforcement learning framework that leverages domain knowledge and uncertainty-aware belief models to enable sample-efficient adaptive sensing and optimal station selection in high-dimensional, low-sample-size Earth observation tasks, as demonstrated by its superior performance in mapping cyanobacterial concentrations in Lake Erie using NASA PACE hyperspectral imagery.
This paper proposes a patient-adaptive transformer framework that combines self-supervised pretraining with patient-specific fine-tuning to achieve high-accuracy short-horizon seizure prediction from EEG recordings, effectively addressing inter-patient variability and complex temporal signal structures.
This paper argues that throughput optimization has evolved into a critical strategic lever for large-scale AI systems, demonstrating through evidence from dataloader frameworks like OVERLORD, memory techniques such as ZeRO-Offload, and compiler-centric tools like Triton-distributed that a holistic, system-level approach is essential to overcome computational bottlenecks and accelerate the development of next-generation foundation models.