Recently, deep neural networks have demonstrated incredible performances, leading to inquiries towards their underlying mechanisms. In this presentation, I will briefly review some most promising mechanisms along three different granularity levels: representation, module, and neuron. I will present some of our lab’s works along each of the granularity levels, and how I consider the future interpretability research will develop.

Dr. Zining Zhu is an Assistant Professor at the Stevens Institute of Technology. He received Ph.D. degree at the University of Toronto and Vector Institute, advised by Frank Rudzicz. He directs the Explainable and Controllable AI lab. He is also affiliated with the Stevens Institute for Artificial Intelligence. He is interested in understanding the mechanisms and abilities of neural network AI systems, and incorporating the findings into controlling the AI systems. In the long term, He looks forward to empowering real-world applications with safe and trustworthy AIs that can collaborate with humans.