If you look at the structure of the lectures, you will find that it is quite consistent with the chapters with the textbook.

I think just reading the textbook or the lecture notes probably is not enough. You probably want to run some codes on some examples by yourself. Perhaps you can run various algorithms within R software or others such as Matlab.

I would consider the field Machine learning something in between Computational Learning Theory and Statistical Learning. Indeed, if you look into Nilsson's book on Introduction to Machine Learning, it kind of provides this flavor. But I honestly think that Computational learning and Statistical learning are quite different. It is more like Discrete Math Vs. Continuous Math, respectively.

Several basic approaches may be useful to our applications:

• Supervised learning: The goal is to derive a correlation Y=C(X). Potential application may be in Delay testing where we usually want to correlate the test results to the golden results (known truth).

• Feature selection: Select a set of features in deriving the correlation. Potential application may be to select test patterns or paths in testing.

• Unsupervised learning: Discover patterns in test data. When there is no golden result, how to discover meaningful patterns in test data.

• Clustering: Identify outliers or abnormal behavior. This may be useful as the front-end to diagnosis.

• Pattern recognition: I think the most powerful application may be in pattern recognition. For example, we may implement a diagnosis tool such that it can automatically identify a problematic layout pattern and also point out all places with similar or same patterns in the design. This will involve using a very efficient pattern recognition tool to process layout image.

In general, we are still in the process of formulating those problems above. We expect some new students can work on these problem in the near future.