Xê-mi-na khoa học
28/06/2022: TS. Lê Đức Trọng (ĐH Công nghệ, ĐHQGHN), Reliable Machine Learning: Concepts, Approaches and Challenges
In supervised learning, researchers often discover underlying patterns from a training dataset using ML algorithms. They expect that these patterns could be exploited to facilitate the prediction task on unseen/future data. This expectation maybe met once there exists an i.i.d distribution among seen and unseen dataset. However, it is not always true in various real-life scenarios. In some cases, the performance on training/validation set is pretty good (> 90%) while it is below the acceptant threshold in the testing/unseen set. Another case, the prediction on future data is first working well, but later becomes unreasonable. These problems raise critical questions on the reliabity of the predictive models, e.g., could we completly be confident on the prediction? How to evaluate their reliablity? How to enhance the their reliablity? If the model is unreliable, it is not applicable in resolving real-life tasks. In this talk, the presenter will give a brief introduction about reliable machine learning including basic concepts, challenges and several practical approaches.
Speaker: TS. Lê Đức Trọng, ĐH Công nghệ, ĐHQGHN
Time: 15:30, Tuesday, June 28, 2022
Venue: Webinar
Dr Le Duc Trong is currently a lecturer at University of Engineering and Technology, Vietnam National University, Hanoi. He received his bachelor degree in Information Technology from VNU-UET (2011); PhD degree in Information Systems from Singapore Management University (2019). His research interest focuses on recommendation systems, reliable machine learning and social/web mining. His papers are published in the proceedings of top-tier conferences such as AAAI, IJCAI, COLING, ACMM. Beside academic activities, he is also a key member in a number of industrial AI projects in various domains funded by companies and organizations.