interpretability – Andrew Fairless, Ph.D.

What I Read: How Machines ‘Grok’ Data

By Andrew Fairless on June 25, 2024April 23, 2024

https://www.quantamagazine.org/how-do-machines-grok-data-20240412 How Do Machines ‘Grok’ Data?Anil Ananthaswamy4/12/24 “By apparently overtraining them, researchers have seen neural networks discover novel solutions to problems.”

What I Read: Generalized Additive Models

By Andrew Fairless on June 13, 2024April 15, 2024

https://ecogambler.netlify.app/blog/interpreting-gams How to interpret and report nonlinear effects from Generalized Additive ModelsNicholas ClarkMarch 21, 2024 “Generalized Additive Models (GAMs)… are helpful for learning arbitrarily complex, nonlinear relationships between predictors andContinue readingWhat I Read: Generalized Additive Models

What I Read: Neural algorithmic reasoning

By Andrew Fairless on January 9, 2024November 7, 2023

https://thegradient.pub/neural-algorithmic-reasoning/ Neural algorithmic reasoningPetar Veličković14.Oct.2023 “…we will talk about classical computation… Think shortest path-finding, sorting, clever ways to break problems down… we will also investigate how to capture such computationContinue readingWhat I Read: Neural algorithmic reasoning

What I Read: Tiny Language Models

By Andrew Fairless on December 11, 2023October 12, 2023

https://www.quantamagazine.org/tiny-language-models-thrive-with-gpt-4-as-a-teacher-20231005/ Tiny Language Models Come of AgeBen Brubaker10/5/23 10:50 AM “To better understand how neural networks learn to simulate writing, researchers trained simpler versions on synthetic children’s stories.”

What I Read: Estimate Token Importance in LLM Prompts

By Andrew Fairless on November 16, 2023October 5, 2023

https://www.watchful.io/blog/a-surprisingly-effective-way-to-estimate-token-importance-in-llm-prompts A Surprisingly Effective Way to Estimate Token Importance in LLM PromptsShayan Mohanty “Prompting is our primary means of instructing these AIs, yet the process of crafting effective prompts hasContinue readingWhat I Read: Estimate Token Importance in LLM Prompts

What I Read: Features Are Important?

By Andrew Fairless on November 2, 2023October 5, 2023

https://www.kdnuggets.com/your-features-are-important-it-doesnt-mean-they-are-good Your Features Are Important? It Doesn’t Mean They Are GoodBy Samuele Mazzanti, Lead Data Scientist at JAKALASeptember 21, 2023 ““Feature Importance” is not enough. You also need to lookContinue readingWhat I Read: Features Are Important?

What I Read: Bonsai Networks, RNNs

By Andrew Fairless on October 3, 2023September 4, 2023

https://cprimozic.net/blog/growing-sparse-computational-graphs-with-rnns/ Growing Bonsai Networks with RNNsCasey Primozic “This writeup introduces what I’m calling Bonsai Networks – extremely sparse computational graphs produced by training and pruning RNNs. They provide an interpretableContinue readingWhat I Read: Bonsai Networks, RNNs

What I Read: Models Memorize or Generalize?

By Andrew Fairless on October 2, 2023September 4, 2023

https://pair.withgoogle.com/explorables/grokking/ Do Machine Learning Models Memorize or Generalize?By Adam Pearce, Asma Ghandeharioun, Nada Hussein, Nithum Thain, Martin Wattenberg and Lucas DixonAugust 2023 “Large language models can certainly seem like theyContinue readingWhat I Read: Models Memorize or Generalize?

What I Read: Shapley Values

By Andrew Fairless on March 2, 2023January 9, 2023

https://www.aidancooper.co.uk/how-shapley-values-work/ How Shapley Values WorkAidan CooperDec 31, 2022 “Shapley values… are machine learning explainability techniques…. we will explore how Shapley values work – not using cryptic formulae, but by wayContinue readingWhat I Read: Shapley Values

What I Read: Biases, Saliency

By Andrew Fairless on February 9, 2023December 16, 2022

https://pair.withgoogle.com/explorables/saliency/ Searching for Unintended Biases With SaliencyAstrid Bertrand, Adam Pearce and Nithum ThainDecember 2022 “If there are spurious correlations in the training data, the model might develop unintended biases thatContinue readingWhat I Read: Biases, Saliency

Tag: interpretability