A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Bryan Y. Lim, Sercan Ö. Arık, Nicolas Loeff, et al.
Introduces the Temporal Fusion Transformer, an attention-based architecture for interpretable multi-horizon time series forecasting that combines high performance with insights into temporal dynamics.
Unknown
Direct3D-S2 introduces spatial sparse attention for scalable 3D generation, enabling practical gigascale 3D synthesis at 1024^3 resolution.
Unknown
The paper derives a solvable attention mechanism that enables precise characterization of neural scaling laws for linear self-attention.
Unknown
SGD-KV compresses LLM KV caches via head-level budget allocation and token-level selection guided by summarization, building on SnapKV-style cumulative attention.
Unknown
KeepKV introduces Electoral Votes for periodic lossless KV cache compression, analyzing compression's impact on attention to enable efficient LLM inference.
Unknown
Introduces a KV cache compression method that estimates attention from the distribution of future queries to reduce memory usage in large language model inference.
Unknown
This study presents a low-rank adaptation method for finetuning large language models, embedding a semantic-aware low-rank adaptation module into attention layers.
Armin Steinhauser
TinyCast is an attention-free zero-shot forecaster with 146K parameters that computes periodicity via a spectral detector, achieving state-of-the-art probabilistic accuracy at its size.
Bo Liu, Qiang Liu
Maglev is a recurrent Transformer architecture with fixed-size memory that generalizes sliding-window attention, trains with a memory consistency loss, and improves validation loss and downstream benchmarks.
Jose Juan Almagro Armenteros, Marco Salvatore, Olof Emanuelsson, et al.
TargetP 2.0 uses deep learning with attention to identify N-terminal sorting signals, revealing that residue 2 strongly influences classification, a feature not previously highlighted.
Unknown
Proposes an efficient sparse attention method for diffusion language models that skips sparse attention in early denoising steps and reuses sparse patterns later.
Unknown
Hcattention achieves extreme KV cache compression via heterogeneous attention computing, combining sparse key selection and value offloading.