A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Supriya V. Mahadevkar, Bharti Khemani, Shruti Patil, et al.
A systematic review of machine learning styles in computer vision, covering supervised, unsupervised, reinforcement, and emerging techniques like zero-shot and self-supervised learning.
Yung-Sung Chuang, Benjamin Cohen-Wang, Shannon Zejiang Shen, et al.
SelfCite introduces a self-supervised method using context ablation to generate reward signals for improving citation quality in LLMs without human annotations.
Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, et al.
ConvNeXt V2 enhances pure convolutional networks with a fully convolutional masked autoencoder and Global Response Normalization for superior self-supervised learning.
Ze Liu, Han Hu, Yutong Lin, et al.
Swin Transformer V2 introduces post-norm, cosine attention, log-spaced position bias, and self-supervised pretraining to scale vision transformers to 3B parameters.
Unknown
Shows self-supervised pre-training can produce general visual features by scaling with curated diverse data, proposing a combined approach for larger models and datasets.
Unknown
Self-supervised Vision Transformers learn explicit semantic segmentation and excel as k-NN classifiers, unlike convolutional networks.
Unknown
SigLIP 2 improves multilingual vision-language encoders with captioning pretraining, self-supervised losses, and online data curation, offering native aspect ratio preservation.
Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, et al.
An LLM learns to autonomously decide when and how to use external APIs, integrating results into text generation via self-supervised fine-tuning.
Unknown
PEGASUS introduces a self-supervised pre-training objective that masks important sentences in documents and generates them as a summary, improving abstractive summarization.
Unknown
This paper integrates knowledge distillation with self-supervised contrastive learning to improve student model performance without requiring labeled data.
Luzhe Huang, Hanlong Chen, Tairan Liu, et al.
GedankenNet eliminates the need for labeled or experimental training data in hologram reconstruction by using a physics-consistency loss and synthetic random images.
Ahmed Elnaggar, Michael Heinzinger, Christian Dallago, et al.
ProtTrans trains protein language models on up to 393 billion amino acids, showing raw embeddings capture biophysical features and outperform state-of-the-art in secondary structure prediction without evolutionary information.