Potential readings

[erictleung]

Potential readings to focus understanding on probability, mathematical statistics, algorithms, etc more deeply. Talk of applications will be secondary.

• Statistical modeling: the two cultures (Brieman, 2001): nice commentary on the tension between statistics (focuses on modeling) and machine learning (focuses on prediction)
• Naught all zeros in sequence count data are the same (Silverman, 2019): understand the reasoning and background behind zero-inflated models in bioinformatics software like DESeq2; good chapter for background reading
• Visualizing data using t-SNE (PDF) (van der Maaten, 2008): popular visualization technique with so many ways explaining what it does and how
• UMAP: uniform manifold approximation and projection for dimension reduction (McInnes, 2018): another popular data visualization method, but not dependent on stochastic sampling
• Kullback-Leibler divergence (PDF): used frequently in bioinformatics, for example (PDF); alternative explanation
• Minimap2: pairwise alignment for nucleotide sequences (Li, 2018): Just review state-of-the-art pairwise alignment algorithms
• Statistical analysis of network data
• Discrete mathematics
• Information theory
• Jeremy Kun's primers: these are excellent blog posts on various topics of interest (e.g., linear algebra, computing theory, probability and statistics)
• Ukkonen's algorithm: really fast algorithm to construct suffix trees; visualization; plain English explanation
• Generative adversarial nets (Goodfellow, 2014): some deep neural network paper
• Quantum machine learning algorithms

[greenstick]

This is great! Another we discussed previously was this one:

• Differential privacy and its applications to validating models, by Cynthia Dwork.

I'll try and think of others and post what I come up with.

[kstevensvt]

I would like to read the 2001 Brieman paper. It looks interesting to me.

[erictleung]

Other potential papers/blogs

• Step away from stepwise: argument against using step-wise building in regression models
• Don't cite the No Free Lunch Theorem
• Pretty much anything from this list

[greenstick]

Nice – interesting about the Wolpert article. I remember reading it and thinking it didn't quite speak to the way 'no free lunch' is used, but then I didn't think much of it.