预训练是研发大语言模型的第一个训练阶段,也是最为重要的一个阶段。Ilya Sutskever 在演讲中直言“预训练(as we know it)将会终结”,暗示需要全新的思路来拓展数据边界。Shital Shah 则在社交媒体上更是指出,真实数据的高质量部分是有限的,继续简单堆砌相似...
PyCodeGPT (CLM): "CERT: Continual Pre-Training on Sketches for Library-Oriented Code Generation" [2022-06] [IJCAI-ECAI 2022] [paper] [repo] PanGu-Coder (CLM): "PanGu-Coder: Program Synthesis with Function-Level Language Modeling" [2022-07] [paper] SantaCoder (FIM): "SantaCoder: don...
A curated list of awesome Deep Reinforcement Learning resources. A list of recent papers regarding deep learning and deep reinforcement learning. A curated list of resources dedicated to reinforcement learning. 子主题 Curated List of Reinforcement Learning Resources for Natural Language Processing. This i...
[Continual-AD]]Continual Learning for Anomaly Detection in Surveillance Videos,CVPR 2020 Worksop. [OGNet]Old is Gold: Redefining the Adversarially Learned One-Class Classifier Training Paradigm,CVPR 2020.code [Any-Shot]Any-Shot Sequential Anomaly Detection in Surveillance Videos,CVPR 2020 workshop. ...
Augmented Partial Mutual Learning with Frame Masking for Video Captioning. [paper] ACMMM-2020 NeurIPS-2020 Prophet Attention: Predicting Attention with Future Attention for Improved Image Captioning. [paper] RATT: Recurrent Attention to Transient Tasks for Continual Image ...
Awesome-Federated-Learning致青**青春 上传25.1 KB 文件格式 zip adversarial-attack-and-defense communication-efficiency computation-efficiency computer-vision continual-learning decentralized-federated-learning distributed-optimization federated-learning hierarchical-federated-learning incentive-mechanism interpretability ...
Forgetting in Continual Learning[Back to top]The goal of continual learning (CL) is to learn on a sequence of tasks without forgetting the knowledge on previous tasks.Links: Task-aware CL | Task-free CL | Online CL | Semi-supervised CL | Few-shot CL | Unsupervised CL | Theoretical ...
Digest One of the main motivations of studying continual learning is that the problem setting allows a model to accrue knowledge from past tasks to learn new tasks more efficiently. However, recent studies suggest that the key metric that continual learning algorithms optimize, reduction in catastroph...
Github: ZhiningLiu1998/awesome-machine-learning-resources用一个多月时间整理了一份几乎涉及机器学习/深度学习一切主题 (学习范式/任务/应用/模型/道德/交叉学科/数据集/框架/教程) 的分类资源列表汇总。由于知…
[2024.03] UniLiDAR: Bridge the domain gap among different LiDARs for continual learning [paper] [2024.03] OccFiner: Offboard Occupancy Refinement with Hybrid Propagation [paper] [2024.03] Unleashing HyDRa: Hybrid Fusion, Depth Consistency and Radar for Unified 3D Perception [paper] [code] [2024.03...