Vanishing Gradients

Hugo speaks with Omoju Miller, a machine learning guru and founder and CEO of Fimio, where she is building 21st century dev tooling. In the past, she was Technical Advisor to the CEO at GitHub, spent time co-leading non-profit investment in Computer Science Education for Google, and served as a volunteer advisor to the Obama administration’s White House Presidential Innovation Fellows.
We need open tools, open data, provenance, and the ability to build fully reproducible, transparent machine learning workflows. With the advent of closed-source, vendor-based APIs and compute becoming a form of gate-keeping, developer tools are at the risk of becoming commoditized and developers becoming consumers.
We’ll talk about how ideas for escaping these burgeoning walled gardens. We’ll dive into
What fully reproducible ML workflows would look like, including git for the workflow build process,
The need for loosely coupled and composable tools that embrace a UNIX-like philosophy,
What a much more scientific toolchain would look like,
What a future open sources commons for Generative AI could look like,
What an open compute ecosystem could look like,
How to create LLMs and tooling so everyone can use them to build production-ready apps,
And much more!
LINKS
The livestream on YouTube (https://www.youtube.com/live/n81PWNsHSMk?si=pgX2hH5xADATdJMu)
Omoju on Twitter (https://twitter.com/omojumiller)
Hugo on Twitter (https://twitter.com/hugobowne)
Vanishing Gradients on Twitter (https://twitter.com/vanishingdata)
Lu.ma Calendar that includes details of Hugo's European Tour for Outerbounds (https://lu.ma/Outerbounds)
Blog post that includes details of Hugo's European Tour for Outerbounds (https://outerbounds.com/blog/ob-on-the-road-2024-h1/)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Hugo speaks with Johno Whitaker, a Data Scientist/AI Researcher doing R&D with answer.ai. His current focus is on generative AI, flitting between different modalities. He also likes teaching and making courses, having worked with both Hugging Face and fast.ai in these capacities.
Johno recently reminded Hugo how hard everything was 10 years ago: “Want to install TensorFlow? Good luck. Need data? Perhaps try ImageNet. But now you can use big models from Hugging Face with hi-res satellite data and do all of this in a Colab notebook. Or think ecology and vision models… or medicine and multimodal models!”
We talk about where we’ve come from regarding tooling and accessibility for foundation models, ML, and AI, where we are, and where we’re going. We’ll delve into
What the Generative AI mindset is, in terms of using atomic building blocks, and how it evolved from both the data science and ML mindsets;
How fast.ai democratized access to deep learning, what successes they had, and what was learned;
The moving parts now required to make GenAI and ML as accessible as possible;
The importance of focusing on UX and the application in the world of generative AI and foundation models;
The skillset and toolkit needed to be an LLM and AI guru;
What they’re up to at answer.ai to democratize LLMs and foundation models.
LINKS
The livestream on YouTube (https://youtube.com/live/hxZX6fBi-W8?feature=share)
Zindi, the largest professional network for data scientists in Africa (https://zindi.africa/)
A new old kind of R&D lab: Announcing Answer.AI (http://www.answer.ai/posts/2023-12-12-launch.html)
Why and how I’m shifting focus to LLMs by Johno Whitaker (https://johnowhitaker.dev/dsc/2023-07-01-why-and-how-im-shifting-focus-to-llms.html)
Applying AI to Immune Cell Networks by Rachel Thomas (https://www.fast.ai/posts/2024-01-23-cytokines/)
Replicate -- a cool place to explore GenAI models, among other things (https://replicate.com/explore)
Hands-On Generative AI with Transformers and Diffusion Models (https://www.oreilly.com/library/view/hands-on-generative-ai/9781098149239/)
Johno on Twitter (https://twitter.com/johnowhitaker)
Hugo on Twitter (https://twitter.com/hugobowne)
Vanishing Gradients on Twitter (https://twitter.com/vanishingdata)
SciPy 2024 CFP (https://www.scipy2024.scipy.org/#CFP)
Escaping Generative AI Walled Gardens with Omoju Miller, a Vanishing Gradients Livestream (https://lu.ma/xonnjqe4)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Hugo speaks with Allen Downey, a curriculum designer at Brilliant, Professor Emeritus at Olin College, and the author of Think Python, Think Bayes, Think Stats, and other computer science and data science books. In 2019-20 he was a Visiting Professor at Harvard University. He previously taught at Wellesley College and Colby College and was a Visiting Scientist at Google. He is also the author of the upcoming book Probably Overthinking It!
They discuss Allen's new book and the key statistical and data skills we all need to navigate an increasingly data-driven and algorithmic world. The goal was to dive deep into the statistical paradoxes and fallacies that get in the way of using data to make informed decisions.
For example, when it was reported in 2021 that “in the United Kingdom, 70-plus percent of the people who die now from COVID are fully vaccinated,” this was correct but the implication was entirely wrong. Their conversation jumps into many such concrete examples to get to the bottom of using data for more than “lies, damned lies, and statistics.” They cover
Information and misinformation around pandemics and the base rate fallacy;
The tools we need to comprehend the small probabilities of high-risk events such as stock market crashes, earthquakes, and more;
The many definitions of algorithmic fairness, why they can't all be met at once, and what we can do about it;
Public health, the need for robust causal inference, and variations on Berkson’s paradox, such as the low-birthweight paradox: an influential paper found that that the mortality rate for children of smokers is lower for low-birthweight babies;
Why none of us are normal in any sense of the word, both in physical and psychological measurements;
The Inspection paradox, which shows up in the criminal justice system and distorts our perception of prison sentences and the risk of repeat offenders.
LINKS
The livestream on YouTube (https://youtube.com/live/G8LulD72kzs?feature=share)
Allen Downey on Github (https://github.com/AllenDowney)
Allen's new book Probably Overthinking It! (https://greenteapress.com/wp/probably-overthinking-it/)
Allen on Twitter (https://twitter.com/AllenDowney)
Prediction-Based Decisions and Fairness: A Catalogue of Choices, Assumptions, and Definitions by Mitchell et al. (https://arxiv.org/abs/1811.07867)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Jeremy Howard (Fast.ai), Shreya Shankar (UC Berkeley), and Hamel Husain (Parlance Labs) join Hugo Bowne-Anderson to talk about how LLMs and OpenAI are changing the worlds of data science, machine learning, and machine learning engineering.
Jeremy Howard (https://twitter.com/jeremyphoward) is co-founder of fast.ai, an ex-Chief Scientist at Kaggle, and creator of the ULMFiT approach on which all modern language models are based. Shreya Shankar (https://twitter.com/sh_reya) is at UC Berkeley, ex Google brain, Facebook, and Viaduct. Hamel Husain (https://twitter.com/HamelHusain) has his own generative AI and LLM consultancy Parlance Labs (https://parlance-labs.com/) and was previously at Outerbounds, Github, and Airbnb.
They talk about
How LLMs shift the nature of the work we do in DS and ML,
How they change the tools we use,
The ways in which they could displace the role of traditional ML (e.g. will we stop using xgboost any time soon?),
How to navigate all the new tools and techniques,
The trade-offs between open and closed models,
Reactions to the recent Open Developer Day and the increasing existential crisis for ML.
LINKS
The panel on YouTube (https://youtube.com/live/MTJHvgJtynU?feature=share)
Hugo and Jeremy's upcoming livestream on what the hell happened recently at OpenAI, among many other things (https://lu.ma/byxyzfrr?utm_source=vg)
Vanishing Gradients on YouTube (https://www.youtube.com/channel/UC_NafIo-Ku2loOLrzm45ABA)
Vanishing Gradients on twitter (https://twitter.com/VanishingData)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Hugo speaks with Hamel Husain, a machine learning engineer who loves building machine learning infrastructure and tools 👷. Hamel leads and contributes to many popular open-source machine learning projects. He also has extensive experience (20+ years) as a machine learning engineer across various industries, including large tech companies like Airbnb and GitHub. At GitHub, he led CodeSearchNet (https://github.com/github/CodeSearchNet), a large language model for semantic search that was a precursor to CoPilot. Hamel is the founder of Parlance-Labs (https://parlance-labs.com/), a research and consultancy focused on LLMs.
They talk about generative AI, large language models, the business value they can generate, and how to get started.
They delve into
Where Hamel is seeing the most business interest in LLMs (spoiler: the answer isn’t only tech);
Common misconceptions about LLMs;
The skills you need to work with LLMs and GenAI models;
Tools and techniques, such as fine-tuning, RAGs, LoRA, hardware, and more!
Vendor APIs vs OSS models.
LINKS
Our upcoming livestream LLMs, OpenAI Dev Day, and the Existential Crisis for Machine Learning Engineering with Jeremy Howard (Fast.ai), Shreya Shankar (UC Berkeley), and Hamel Husain (Parlance Labs): Sign up for free! (https://lu.ma/m81oepqe/utm_source=vghh)
Our recent livestream Data and DevOps Tools for Evaluating and Productionizing LLMs (https://youtube.com/live/B_DMMlDuJB0) with Hamel and Emil Sedgh, Lead AI engineer at Rechat -- in it, we showcase an actual industrial use case that Hamel and Emil are working on with Rechat, a real estate CRM, taking you through LLM workflows and tools.
Extended Guide: Instruction-tune Llama 2 (https://www.philschmid.de/instruction-tune-llama-2) by Philipp Schmid
The livestream recoding of this episode! (https://youtube.com/live/l7jJhL9geZQ?feature=share)
Hamel on twitter (https://twitter.com/HamelHusain)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Hugo speaks with Chris Wiggins (Columbia, NYTimes) and Matthew Jones (Princeton) about their recent book How Data Happened, and the Columbia course it expands upon, data: past, present, and future.
Chris is an associate professor of applied mathematics at Columbia University and the New York Times’ chief data scientist, and Matthew is a professor of history at Princeton University and former Guggenheim Fellow.
From facial recognition to automated decision systems that inform who gets loans and who receives bail, we all now move through a world determined by data-empowered algorithms. These technologies didn’t just appear: they are part of a history that goes back centuries, from the census enshrined in the US Constitution to the birth of eugenics in Victorian Britain to the development of Google search.
DJ Patil, former U.S. Chief Data Scientist, said of the book "This is the first comprehensive look at the history of data and how power has played a critical role in shaping the history. It’s a must read for any data scientist about how we got here and what we need to do to ensure that data works for everyone."
If you’re a data scientist, machine learning engineer, or work with data in any way, it’s increasingly important to know more about the history and future of the work that you do and understand how your work impacts society and the world.
Among other things, they'll delve into
* the history of human use of data;
* how data are used to reveal insight and support decisions;
* how data and data-powered algorithms shape, constrain, and manipulate our commercial, civic, and personal transactions and experiences; and
* how exploration and analysis of data have become part of our logic and rhetoric of communication and persuasion.
You can also sign up for our next livestreamed podcast recording here (https://www.eventbrite.com/e/data-science-past-present-and-future-tickets-695643357007?aff=kjvg)!
LINKS
How Data Happened, the book! (https://wwnorton.com/books/how-data-happened)
data: past, present, and future, the course (https://data-ppf.github.io/)
Race After Technology, by Ruha Benjamin (https://www.ruhabenjamin.com/race-after-technology)
The problem with metrics is a big problem for AI by Rachel Thomas (https://www.ruhabenjamin.com/race-after-technology)
Vanishing Gradients on YouTube (https://www.youtube.com/channel/UC_NafIo-Ku2loOLrzm45ABA)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Hugo speaks with Katharine Jarmul about privacy and security in data science and machine learning. Katharine is a Principal Data Scientist at Thoughtworks Germany focusing on privacy, ethics, and security for data science workflows. Previously, she has held numerous roles at large companies and startups in the US and Germany, implementing data processing and machine learning systems with a focus on reliability, testability, privacy, and security.
In this episode, Hugo and Katharine talk about
What data privacy and security are, what they aren’t and the differences between them (hopefully dispelling common misconceptions along the way!);
Why you should care about them (hint: the answers will involve regulatory, ethical, risk, and organizational concerns);
Data governance, anonymization techniques, and privacy in data pipelines;
Privacy attacks!
The state of the art in privacy-aware machine learning and data science, including federated learning;
What you need to know about the current state of regulation, including GDPR and CCPA…
And much more, all the while grounding our conversation in real-world examples from data science, machine learning, business, and life!
You can also sign up for our next livestreamed podcast recording here (https://lu.ma/4b5xalpz)!
LINKS
Win a copy of Practical Data Privacy, Katharine's new book! (https://forms.gle/wkF92vyvjfZLM6qt8)
Katharine on twitter (https://twitter.com/kjam)
Vanishing Gradients on YouTube (https://www.youtube.com/channel/UC_NafIo-Ku2loOLrzm45ABA)
Probably Private, a newsletter for privacy and data science enthusiasts (https://probablyprivate.com/)
Probably Private on YouTube (https://www.youtube.com/@ProbablyPrivate)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Hugo speaks with Eric Ma about Research Data Science in Biotech. Eric leads the Research team in the Data Science and Artificial Intelligence group at Moderna Therapeutics. Prior to that, he was part of a special ops data science team at the Novartis Institutes for Biomedical Research's Informatics department.
In this episode, Hugo and Eric talk about
What tools and techniques they use for drug discovery (such as mRNA vaccines and medicines);
The importance of machine learning, deep learning, and Bayesian inference;
How to think more generally about such high-dimensional, multi-objective optimization problems;
The importance of open-source software and Python;
Institutional and cultural questions, including hiring and the trade-offs between being an individual contributor and a manager;
How they’re approaching accelerating discovery science to the speed of thought using computation, data science, statistics, and ML.
And as always, much, much more!
LINKS
Eric's website (https://ericmjl.github.io/)
Eric on twitter (https://twitter.com/ericmjl)
Vanishing Gradients on YouTube (https://www.youtube.com/channel/UC_NafIo-Ku2loOLrzm45ABA)
Cell Biology by the Numbers by Ron Milo and Rob Phillips (http://book.bionumbers.org/)
Eric's JAX tutorials at PyCon (https://youtu.be/ztthQJQFe20) and SciPy (https://youtu.be/DmR36wtel4Y)
Eric's blog post on Hiring data scientists at Moderna! (https://ericmjl.github.io/blog/2021/8/26/hiring-data-scientists-at-moderna-2021/)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Hugo speaks with Tanya Cashorali, a data scientist and consultant that helps businesses get the most out of data, about what end-to-end data science looks like across many industries, such as retail, defense, biotech, and sports, including
scoping out projects,
figuring out the correct questions to ask,
how projects can change,
delivering on the promise,
the importance of rapid prototyping,
what it means to put models in production, and
how to measure success.
And much more, all the while grounding their conversation in real-world examples from data science, business, and life.
In a world where most organizations think they need AI and yet 10-15% of data science actually involves model building, it’s time to get real about how data science and machine learning actually deliver value!
LINKS
Tanya on Twitter (https://twitter.com/tanyacash21)
Vanishing Gradients on YouTube (https://www.youtube.com/channel/UC_NafIo-Ku2loOLrzm45ABA)
Saving millions with a Shiny app | Data Science Hangout with Tanya Cashorali (https://youtu.be/qdAroyFRFCg)
Our next livestream: Research Data Science in Biotech with Eric Ma (https://www.eventbrite.com/e/research-data-science-in-biotech-tickets-550400882857?aff=fs)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe

Hugo speaks with JD Long, agricultural economist, quant, and stochastic modeler, about decision making under uncertainty and how we can use our knowledge of risk, uncertainty, probabilistic thinking, causal inference, and more to help us use data science and machine learning to make better decisions in an uncertain world.
This is part 2 of a two part conversation in which we delve into decision making under uncertainty. Feel free to check out part 1 here (https://vanishinggradients.fireside.fm/15) but this episode should also stand alone.
Why am I speaking to JD about all of this? Because not only is he a wild conversationalist with a real knack for explaining hard to grok concepts with illustrative examples and useful stories, but he has worked for many years in re-insurance, that’s right, not insurance but re-insurance – these are the people who insure the insurers so if anyone can actually tell us about risk and uncertainty in decision making, it’s him!
In part 1, we discussed risk, uncertainty, probabilistic thinking, and simulation, all with a view towards improving decision making.
In this, part 2, we discuss the ins and outs of decision making under uncertainty, including
How data science can be more tightly coupled with the decision function in organisations;
Some common mistakes and failure modes of making decisions under uncertainty;
Heuristics for principled decision-making in data science;
The intersection of model building, storytelling, and cognitive biases to keep in mind;
As JD says, and I paraphrase, “You may think you train your models, but your models are really training you.”
Links
Vanishing Gradients' new YouTube channel! (https://www.youtube.com/channel/UC_NafIo-Ku2loOLrzm45ABA)
JD on twitter (https://twitter.com/CMastication)
Executive Data Science, episode 5 of Vanishing Gradients, in which Jim Savage and Hugo talk through decision making and why you should always be integrating your loss function over your posterior (https://vanishinggradients.fireside.fm/5)
Fooled by Randomness by Nassim Taleb (https://en.wikipedia.org/wiki/Fooled_by_Randomness)
Superforecasting: The Art and Science of Prediction Philip E. Tetlock and Dan Gardner (https://en.wikipedia.org/wiki/Superforecasting:_The_Art_and_Science_of_Prediction)
Thinking in Bets by Annie Duke (https://www.penguin.com.au/books/thinking-in-bets-9780735216372)
The Signal and the Noise: Why So Many Predictions Fail by Nate Silver (https://en.wikipedia.org/wiki/The_Signal_and_the_Noise)
Thinking, Fast and Slow by Daniel Kahneman (https://en.wikipedia.org/wiki/Thinking,_Fast_and_Slow)

Get full access to Vanishing Gradients at hugobowne.substack.com/subscribe