A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Kamal Berahmand, Fatemeh Daneshfar, Elaheh Sadat Salehi, et al.
A comprehensive survey of autoencoders, providing a taxonomy based on structures and principles, reviewing applications across multiple domains, and outlining limitations and future directions.
Unknown
This paper proposes a method to enable diffusion transformers to operate effectively within latent spaces by integrating representation autoencoders.
Unknown
SPARE is a training-free method using pre-trained sparse auto-encoders to steer knowledge selection in LLMs by editing internal representations.
Riccardo Miotto, Li Li, Brian Kidd, et al.
Unsupervised deep feature learning with stacked denoising autoencoders on EHRs of ~700k patients yields a general-purpose patient representation that significantly improves predictive modeling for 78 diseases.
Unknown
DocLLM extends LLMs to reason over visual documents using textual semantics and spatial layout, avoiding expensive image encoders.
Unknown
AIM V2 introduces a family of open vision encoders from 300M to 3B parameters, pre-trained with a multimodal autoregressive approach that jointly generates image patches and text tokens using a causal decoder.
Unknown
SigLIP 2 improves multilingual vision-language encoders with captioning pretraining, self-supervised losses, and online data curation, offering native aspect ratio preservation.
Andreas Koukounas, Georgios Mastrapas, Michael Günther, et al.
Contrastively fine-tunes T5 encoders on curated data to distinguish negations from confirmations.
Unknown
Eagle 2 introduces a data-centric post-training strategy and a tiled mixture of vision encoders to build high-performing vision-language models.
Unknown
Eagle systematically explores the design space for multimodal LLMs using a mixture of vision encoders and resolutions, revealing key principles for effective fusion.
Weihan Wang, Qingsong Lv, Wenmeng Yu, et al.
CogVLM bridges the gap between frozen pretrained language models and image encoders using a trainable visual expert module in attention and FFN layers.
Unknown
BLIP-2 introduces Q-Former, a lightweight transformer module that bridges frozen image encoders and frozen LLMs for efficient vision-language pre-training.