Skip to main content

Posters

2024

Rethinking Model-based, Policy-based, and Value-based Reinforcement Learning via the Lens of Representation Complexity
·1788 words·9 mins· loading · loading
Machine Learning Reinforcement Learning 🏢 Peking University
Reinforcement learning paradigms exhibit a representation complexity hierarchy: models are easiest, then policies, and value functions are hardest to approximate.
Rethinking Misalignment in Vision-Language Model Adaptation from a Causal Perspective
·2085 words·10 mins· loading · loading
Multimodal Learning Vision-Language Models 🏢 Institute of Software Chinese Academy of Sciences
Vision-language model adaptation struggles with misalignment; this paper introduces Causality-Guided Semantic Decoupling and Classification (CDC) to mitigate this, boosting performance.
Rethinking Memory and Communication Costs for Efficient Data Parallel Training of Large Language Models
·2992 words·15 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Ant Group
PaRO boosts LLM training speed by up to 266% through refined model state partitioning and optimized communication.
Rethinking LLM Memorization through the Lens of Adversarial Compression
·2014 words·10 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Carnegie Mellon University
Researchers propose Adversarial Compression Ratio (ACR) to assess LLM memorization, offering an adversarial, flexible, and computationally efficient method for monitoring data misuse and compliance.
Rethinking Inverse Reinforcement Learning: from Data Alignment to Task Alignment
·2568 words·13 mins· loading · loading
AI Generated Machine Learning Reinforcement Learning 🏢 Boston University
PAGAR: a novel semi-supervised IRL framework prioritizing task alignment over data alignment, leveraging expert demonstrations as weak supervision to derive task-aligned reward functions for improved …
Rethinking Imbalance in Image Super-Resolution for Efficient Inference
·2134 words·11 mins· loading · loading
Computer Vision Image Generation 🏢 Harbin Institute of Technology
WBSR: A novel framework for efficient image super-resolution that tackles data and model imbalances for superior performance and approximately a 34% reduction in computational cost.
Rethinking Human Evaluation Protocol for Text-to-Video Models: Enhancing Reliability, Reproducibility, and Practicality
·3943 words·19 mins· loading · loading
AI Generated Multimodal Learning Vision-Language Models 🏢 University of California San Diego
This paper presents Text-to-Video Human Evaluation (T2VHE), a new protocol for evaluating text-to-video models, improving reliability, reproducibility, and practicality.
Rethinking Fourier Transform from A Basis Functions Perspective for Long-term Time Series Forecasting
·1994 words·10 mins· loading · loading
Machine Learning Deep Learning 🏢 School of Computing, Macquarie University
Revolutionizing long-term time series forecasting, a new Fourier Basis Mapping method enhances accuracy by precisely interpreting frequency coefficients and considering time-frequency relationships, a…
Rethinking Deep Thinking: Stable Learning of Algorithms using Lipschitz Constraints
·2740 words·13 mins· loading · loading
Machine Learning Deep Learning 🏢 University of Southampton
Stable algorithm learning achieved by Deep Thinking networks with Lipschitz Constraints, ensuring convergence and better extrapolation to complex problems.
Rethinking Decoders for Transformer-based Semantic Segmentation: Compression is All You Need
·2306 words·11 mins· loading · loading
Computer Vision Image Segmentation 🏢 Beijing University of Posts and Telecommunications
DEPICT: A new white-box decoder for Transformer-based semantic segmentation, achieving better performance with fewer parameters by leveraging the principle of compression and connecting Transformer de…
RestoreAgent: Autonomous Image Restoration Agent via Multimodal Large Language Models
·2570 words·13 mins· loading · loading
Computer Vision Image Generation 🏢 Hong Kong University of Science and Technology
RestoreAgent, an AI-powered image restoration agent, autonomously identifies and corrects multiple image degradations, exceeding human expert performance.
ReST-MCTS*: LLM Self-Training via Process Reward Guided Tree Search
·2799 words·14 mins· loading · loading
AI Generated Natural Language Processing Large Language Models 🏢 Tsinghua University
ReST-MCTS*: A novel LLM self-training method using process reward guided tree search, outperforming existing methods by generating higher-quality reasoning traces for improved model accuracy.
Resource-Aware Federated Self-Supervised Learning with Global Class Representations
·2832 words·14 mins· loading · loading
AI Generated Machine Learning Self-Supervised Learning 🏢 Shandong University
FedMKD: A multi-teacher framework for federated self-supervised learning, enabling global class representations even with diverse client models and skewed data distributions.
Reshuffling Resampling Splits Can Improve Generalization of Hyperparameter Optimization
·4058 words·20 mins· loading · loading
AI Theory Optimization 🏢 Munich Center for Machine Learning (MCML)
Reshuffling data splits during hyperparameter optimization surprisingly improves model generalization, offering a computationally cheaper alternative to standard methods.
Resfusion: Denoising Diffusion Probabilistic Models for Image Restoration Based on Prior Residual Noise
·2678 words·13 mins· loading · loading
Computer Vision Image Generation 🏢 College of Computer Science, Nankai University
Resfusion, a novel framework, accelerates image restoration by integrating residual noise into the diffusion process, achieving superior results with fewer steps.
Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe
·5026 words·24 mins· loading · loading
AI Generated Natural Language Processing Large Language Models 🏢 University of Cambridge
This research unveils a compute-optimal recipe for fine-tuning language models into high-quality text embedding models, offering practical guidance and scaling laws for resource-constrained settings.
Reprogramming Pretrained Target-Specific Diffusion Models for Dual-Target Drug Design
·1732 words·9 mins· loading · loading
AI Applications Healthcare 🏢 Tsinghua University
AI-powered dual-target drug design is revolutionized by repurposing pretrained diffusion models, achieving zero-shot transfer learning and outperforming existing methods.
Representation Noising: A Defence Mechanism Against Harmful Finetuning
·3502 words·17 mins· loading · loading
Natural Language Processing Large Language Models 🏢 Dalhousie University
RepNoise: a novel defense against harmful fine-tuning of LLMs by removing information about harmful representations, generalizing across different harmful tasks, and maintaining LLM capabilities.
Replicable Uniformity Testing
·268 words·2 mins· loading · loading
AI Generated AI Theory Optimization 🏢 UC San Diego
This paper presents the first replicable uniformity tester with nearly linear dependence on the replicability parameter, enhancing the reliability of scientific studies using distribution testing algo…
Replicability in Learning: Geometric Partitions and KKM-Sperner Lemma
·301 words·2 mins· loading · loading
AI Theory Optimization 🏢 Sandia National Laboratories
This paper reveals near-optimal relationships between geometric partitions and replicability in machine learning, establishing the optimality of existing algorithms and introducing a new neighborhood …
Buy Me A Coffee