update research ovws

_notes/stat/causal_inference.md

@@ -23,6 +23,7 @@ category: stat
 # basic causal inference
 
 - *confounding* - difference between groups other than the treatment which affects the response
+- [bradford hill criteria](https://en.wikipedia.org/wiki/Bradford_Hill_criteria) - some simple criteria for establishing causality (e.g. strength, consistency, specificity)
 - 3 frameworks
   1. neyman-rubin model: $Y_i = T_i a_i + (1-T_i) b_i$
     - $\hat{ate} = \hat{a}_A - \hat{b}_B$

_research_ovws/ovw_complexity.md

@@ -46,6 +46,9 @@ category: research
     - algorithmic / mdl
     - natural complexity (e.g. physical complexity)
       - ![300px-Complexity_figure1](assets/300px-Complexity_figure1.jpg)
+- [quanta article](https://www.quantamagazine.org/computer-science-and-biology-explore-algorithmic-evolution-20181129/?fbclid=IwAR0rSImplo7lLM0kEYHrHttx8qUimB-482dI9IFxY6dvx0CFeEIqzGuir_w)
+  - "the probability of producing some types of outputs is far greater when randomness operates at the level of the program describing it rather than at the level of the output itself"
+  - "they [recently reported in *Royal Society Open Science*](http://rsos.royalsocietypublishing.org/content/5/8/180399) that, compared to statistically random mutations, this mutational bias caused the networks to evolve toward solutions significantly faster."
 
 # minimum description length
 

_research_ovws/ovw_interp.md

@@ -11,6 +11,10 @@ category: research
 
 # misc new papers
 
+- [Beyond Sparsity: Tree Regularization of Deep Models for Interpretability](https://arxiv.org/pdf/1711.06178.pdf)
+  - regularize so that deep model can be closely modeled by tree w/ few nodes
+- [THE CONVOLUTIONAL TSETLIN MACHINE](https://arxiv.org/pdf/1905.09688.pdf)
+  - [The Tsetlin Machine](https://arxiv.org/pdf/1804.01508.pdf)
 - [explaining image classifiers by counterfactual generation](https://arxiv.org/pdf/1807.08024.pdf) 
   - generate changes (e.g. with GAN in-filling) and see if pred actually changes
 - [ConvNets and ImageNet Beyond Accuracy: Understanding Mistakes and Uncovering Biases](https://arxiv.org/abs/1711.11443)
@@ -71,8 +75,7 @@ category: research
   - could also just use specific filters for specific classes
 - [Manipulating and Measuring Model Interpretability](https://arxiv.org/abs/1802.07810)
 - local
-  - [grad-cam++](https://arxiv.org/abs/1710.11063) 
-  - [What made you do this? Understanding black-box decisions with sufficient input subsets](https://arxiv.org/pdf/1810.03805.pdf)
+  - [grad-cam++](https://arxiv.org/abs/1710.11063)
 - importance scores
   - [Variable Importance Clouds: A Way to Explore Variable Importance for the Set of Good Models](https://arxiv.org/pdf/1901.03209.pdf) 
   - [Permutation tests in general](http://arxiv.org/abs/1801.01489): Fisher, Aaron, Cynthia Rudin, and Francesca Dominici. 2018. “Model Class Reliance: Variable Importance Measures for any Machine Learning Model Class, from the ‘Rashomon’ Perspective.”
@@ -298,8 +301,15 @@ category: research
     - this includes sets of different sizes
 - shapley sampling value - sample instead of exactly computing
   - quantitative input influence is similar to this...
+- [A Unified Approach to Interpreting Model Predictions](http://papers.nips.cc/paper/7062-a-unified-approach-to-interpreting-model-predicti)
+  - 3 properties
+    - local accuracy - basically, explanation scores sum to original prediction
+    - missingness - features with $x'_i=0$ have 0 impact
+    - consistency - if a model changes so that some simplified input’s contribution increases or stays the same regardless of the other inputs, that input’s attribution should not decrease.
+  - ![Screen Shot 2019-05-28 at 10.23.49 PM](../../Desktop/Screen Shot 2019-05-28 at 10.23.49 PM.png)
 
 ## example-based explanations
+
 - influential instances - want to find important data points
   - deletion diagnostics - delete a point and see how much it changed
 
@@ -339,6 +349,9 @@ category: research
    - Partial Dependence Plots: “Let me show you what the model predicts on average when each data instance has the value v for that feature. I ignore whether the value v makes sense for all data instances.” 
 - M-Plots: “Let me show you what the model predicts on average for data instances that have values close to v for that feature. The effect could be due to that feature, but also due to correlated features.” 
   - ALE plots: “Let me show you how the model predictions change in a small “window” of the feature around v for data instances in that window.” 
+- [What made you do this? Understanding black-box decisions with sufficient input subsets](https://arxiv.org/pdf/1810.03805.pdf)
+   - want to find smallest subsets of features which can produce the prediction
+      - other features are masked or imputed
 
 
 

readme.md

@@ -4,39 +4,35 @@
 
 ## slides
 
-- the [_slides](_slides) folder contains source for many presentations, including [ml slides](https://csinva.github.io/pres/189/#/) from teaching intro machine learning at berkeley:
+- the [_slides](_slides) folder contains source for presentations, including [ml slides](https://csinva.github.io/pres/189/#/) from teaching intro machine learning at berkeley
 
 ![](assets/img/pres_demo.gif)
 
 ## research overviews
 
-- notes on different research areas in the [research_ovws](_notes/research_ovws) folder (e.g. interpretability, complexity) with links to recent papers
+- overviews and summaries of recent papers in different research areas in the [research_ovws](_notes/research_ovws) folder (e.g. [interpretable ml](https://github.com/csinva/csinva.github.io/blob/master/_research_ovws/ovw_interp.md),  [theoretical ml](https://github.com/csinva/csinva.github.io/blob/master/_research_ovws/ovw_dl_theory.md),  [complexity in ml](https://github.com/csinva/csinva.github.io/blob/master/_research_ovws/ovw_complexity.md), [scattering transform](https://github.com/csinva/csinva.github.io/blob/master/_research_ovws/ovw_scat.md), [dl in neuroscience](https://github.com/csinva/csinva.github.io/blob/master/_research_ovws/ovw_dl_for_neuro.md))
 
 ## notes
 
 - the [_notes](_notes) folder contains markdown notes for many different classes and areas between computer science, statistics, and neuroscience
 
-## blog posts
-
-
-| [interpretability](https://csinva.github.io/blog/interpretability) | [connectomics](https://csinva.github.io/blog/connectomics) |
-| ------------------------------------------------------------ | ------------------------------------------------------------ |
-|            ![](assets/img/alexnet.png)                                                   | ![](assets/img/400by31Normalized.gif)                        |
-
 
+## code
 
+- links to research code, such as these repos for interpretable machine learning (e.g. [hierarchical-dnn-interpretations](https://github.com/csinva/hierarchical_dnn_interpretations), [disentangled-attribution-curves](https://github.com/csinva/disentangled_attribution_curves)) and miscellaneous repos for data science (e.g. [pytorch-pretrained-gans](https://github.com/csinva/pytorch_gan_pretrained)), which contains code for simple pre-trained gans
 
-## code
+![](assets/img/acd_intro.png)
 
-- links to code, such as these repos ([acd](https://github.com/csinva/hierarchical_dnn_interpretations), [dac](https://github.com/csinva/disentangled_attribution_curves)) for interpretable machine learning and [this repo](https://github.com/csinva/pytorch_gan_pretrained), which contains code for simple pre-trained gans
+## blog posts
 
-| generated samples                                            | data samples     |
-| ------------------------------------------------------------ | ----------------------------------------------------- |
-| ![fake_images-300](assets/img/samples/fake_samples_epoch_199.png) | ![real_images](assets/img/samples/real_samples.png) |
 
+| [interpretability](https://csinva.github.io/blog/interpretability) | [connectomics](https://csinva.github.io/blog/connectomics) |
+| ------------------------------------------------------------ | ------------------------------------------------------------ |
+|            ![](assets/img/alexnet.png)                                                   | ![](assets/img/400by31Normalized.gif)                        |
 
 
 # reference
 
-- uses theme from [here](http://kirbyt.github.io/timeline-jekyll-theme)
+- for updates, see [@chandan_singh96](https://twitter.com/chandan_singh96)
+- uses [timeline theme](http://kirbyt.github.io/timeline-jekyll-theme)
 - uses [particles.js](https://vincentgarreau.com/particles.js/)