What is the future of AI / ML ?

I believe there will be significant breakthroughs in the following four aspects of AI and machine learning (ML) in the coming decade. I have worked and will continue to work on these topics.

1) Automated ML. Machine learning is accessible to the experts, while AutoML is accessible to non-technical users. Future works on AutoML might include neural architecture search in CV/NLP/DM, hyper-parameter search, and metric learning of neural architecture.

2) Trustable ML. I define the trustable ML into two sub-topics, including interpretable neural networks and robust neural networks. In particular, I am working on these two aspects passionately.

3) Unsupervised ML. As we know, self-supervised learning has obtained slightly lower accuracies on ImageNet (but with significantly higher computational cost) than the supervised learning. Hence, unsupervised learning might match the performance of supervised learning in the coming two years. Anyone who is the first to make it should be highly rated.

4) Causal learning and brain-like learning.

返回博客目录Return to all Blogs
返回主页Return to homepage