Interactive Debugging and Testing Support for Deep Learning
summary
An interactive debugger called DeepSeer transforms RNN internals into a Finite State Machine so programmers can step through a state trace and inspect words tied to each state. Applied to sentiment RNNs it surfaces a clear root cause — e.g. “quarters” appears 28 times in training with 27 positive labels — and highlights learned spurious correlations. A user study with 28 CS students comparing DeepSeer to Line yields more correct root‑cause insights and more error‑inducing keywords found.
AI-written summary · about
chapters (6)
- 0:00The challenges of AI
- 1:15Why deep learning is a box
- 2:42Introducing DeepSeer
- 3:06Tool interface and usage
- 4:02Debugging example and study
- 5:31Key takeaways and conclusion
related talks
53mTensorflow, deep learning and modern RNN architectures, without a PhD
39mTackling the Limits of Deep Learning
37mDeep Recommender Systems
10mDeepSI: Interactive Deep Learning for Semantic Interaction
30mOutwitting Deep Learning Models
43mDebugging AI
59mDeep Neural Networks for Video Applications
30mSemantic Search and Neural Information Retrieval