🏢 Nanjing University

This paper introduces a novel universal online convex optimization algorithm needing only one projection per round, achieving optimal regret bounds for various function types, including general convex…

START, a novel SSM-based architecture with saliency-driven token-aware transformation, achieves state-of-the-art domain generalization performance with efficient linear complexity.

SOFTS: An efficient MLP-based model for multivariate time series forecasting using a novel STAR module for efficient channel interaction.

SCaR refines skill chaining for long-horizon robotic manipulation via dual regularization, achieving higher success rates and robustness.

Reinforcement learning refines existing macro placements, enhancing chip design by improving power, performance, and area (PPA) metrics and integrating the often-overlooked metric of regularity.

Prism: a novel framework disentangles perception and reasoning in Vision-Language Models (VLMs) for improved model assessment and efficient VLM development.

Researchers developed Policy Learning from tutorial Books (PLfB), a novel method that trains AI agents using knowledge from tutorial books instead of relying solely on real-world data.

Researchers achieve optimal regret bounds in online composite optimization under stochastic and adversarial settings using a novel optimistic composite mirror descent algorithm and a universal strateg…

Learnware systems enable model reuse; this paper proves RKME specifications protect developers’ training data while enabling effective model identification.

Neuro-symbolic framework generates high-quality mathematical datasets, enhancing LLMs’ mathematical reasoning capabilities and surpassing state-of-the-art counterparts.

MCTS-transfer: Iteratively refining Bayesian optimization via Monte Carlo tree search for efficient black-box optimization using transfer learning.

Meta-DT: Offline meta-RL masters unseen tasks via conditional sequence modeling and world model disentanglement, showcasing superior few-shot and zero-shot generalization.

IODA achieves efficient one-shot domain adaptation for super-resolution using a novel instance-guided strategy and image-level domain alignment, significantly improving performance with limited target…

FlowDCN: A purely convolutional generative model achieves state-of-the-art image generation speed and quality at arbitrary resolutions, surpassing transformer-based models.

Deep neural network training reveals asymmetric loss valleys, impacting model fusion and federated learning; sign consistency between noise and convergence is key.

Recurrent off-policy RL, while robust, suffers from training instability. RESEL, a novel algorithm, solves this by using a context-encoder-specific learning rate, significantly improving stability an…

EASI: Evolutionary Adversarial Simulator Identification bridges the reality gap in robotics by using GAN and ES to find optimal simulator parameters, enabling seamless sim-to-real transfer with minima…

Multi-Path Aggregation (MPA) achieves comparable performance to state-of-the-art methods in multi-task image coding, by unifying feature representations with a novel all-in-one architecture and a two-…

Adaptive STORM achieves optimal convergence rates for stochastic optimization of non-convex functions under weaker assumptions, eliminating the need for bounded gradients or function values and removi…

A novel universal online learning algorithm achieves optimal gradient-variation regret across diverse function curvatures, boasting efficiency with only one gradient query per round.