Proposes an end-to-end differentiable Transformer Neural Process framework for meta Bayesian optimization, leveraging reinforcement learning to improve sample efficiency.

Alexandre Maraval, Matthieu Zimmer, Antoine Grosnit et al.

Plasticity injection enhances neural network adaptability in deep RL, boosting performance by 20% on Atari without increasing trainable parameters.

Evgenii Nikishin, Junhyuk Oh, Georg Ostrovski et al.

SBMI method infers joint probability distributions of model components and parameters using neural networks.

Cornelius Schröder, Jakob H. Macke

VIPER leverages pretrained video prediction models as reward signals, enabling complex behavior learning without task-specific rewards.

Alejandro Escontrela, Ademi Adeniji, Wilson Yan et al.

QLoRA combines 4-bit quantization, LoRA, and paging to finetune 65B models on a single GPU, achieving near ChatGPT performance.

Tim Dettmers, Artidoro Pagnoni, Ari Holtzman et al.

PEQA method enables sub-4-bit LLM fine-tuning by updating quantization scales, significantly reducing memory demands.

Jeonghoon Kim, Jung Hyun Lee, Sungdong Kim et al.

Introduces DDPO, a reinforcement learning framework that directly optimizes diffusion models for target-specific objectives, improving image quality and alignment.

Kevin Black, Michael Janner, Yilun Du et al.

Diff-Pruning uses Taylor expansion to cut 50% FLOPs in diffusion models, with only 10-20% training cost, maintaining quality.

Gongfan Fang, Xinyin Ma, Xinchao Wang

Smoothing the loss landscape enables online SGD to achieve the optimal sample complexity of n ≥ d^{k^*/2} for learning single index models, matching the CSQ lower bound.

Alex Damian, Eshaan Nichani, Rong Ge et al.

Introduces Intention module based on least squares solution, with complexity matching attention, enhancing model expressiveness beyond attention's limits.

Marta Garnelo, Wojciech Marian Czarnecki

FrugalGPT employs adaptive model cascade to reduce LLM inference costs by up to 98%, matching or surpassing GPT-4 accuracy.

Lingjiao Chen, Matei Zaharia, James Zou

GeoLDM integrates geometric autoencoders with latent diffusion, boosting 3D molecule generation with up to 7% higher validity.

Minkai Xu, Alexander Powers, Ron Dror et al.

Proposes CoMOT with meta-OT and GumMS for fast, generalizable fair re-ranking, outperforming BvND in speed and storage.

Andrés Hoyos-Idrobo

TGNN combines message passing and graph kernel modules with consistency loss for semi-supervised graph classification, outperforming baselines.

Wei Ju, Xiao Luo, Meng Qu et al.

VALOR model achieves tri-modal learning through MGA and MGC tasks for vision, audio, and language.

Jing Liu, Sihan Chen, Xingjian He et al.

This survey systematically reviews deep graph representation learning architectures, paradigms, and applications, highlighting GNN innovations and future challenges.

Wei Ju, Zheng Fang, Yiyang Gu et al.

Combining Chinchilla scaling laws and μP, trained models from 111M to 13B parameters achieving state-of-the-art compute efficiency.

Nolan Dey, Gurpreet Gosal, Zhiming et al.

BloombergGPT is a 50B-parameter finance-specific LLM trained on 700B tokens, combining financial and general data for superior performance.

Shijie Wu, Ozan Irsoy, Steven Lu et al.

Proposes Diffusion Classifier, leveraging diffusion model density estimates for zero-shot image classification, outperforming traditional discriminative methods.

Alexander C. Li, Mihir Prabhudesai, Shivam Duggal et al.

Proposes a model-based explanation framework for concept drift, leveraging spatial feature changes and explanation tools like LIME and SHAP.

Fabian Hinder, Valerie Vaquet, Johannes Brinkrolf et al.