On this episode of Learning from Machine Learning, I had the privilege of speaking with Maxime Labonne, Head of Post-Training at Liquid AI. We traced his journey from cybersecurity to the cutting edge of model architecture. Maxime shared how the future of AI isn't just about making models bigger—it's about making them smarter and more efficient.

Maxime's work demonstrates that challenging established paradigms requires taking steps backward to leap forward. His framework for data quality—accuracy, diversity, and complexity—offers a blueprint for anyone working with machine learning systems.

Most importantly, Maxime's perspective on learning itself—treating knowledge acquisition like training data exposure—reminds us that growth comes from diverse, high-quality experiences across different contexts. Whether you're training a model or developing yourself, the principles remain remarkably similar.

Thank you for listening. Be sure to subscribe and share with a friend or colleague. Until next time... keep on learning.

00:46 Introduction and Maxime's Background

01:47 Journey from Cybersecurity to Machine Learning

03:30 The Fascination with AI and Cyber Attacks

06:15 Transitioning to Post-Training at Liquid AI

08:17 Liquid AI's Vision and Mission

10:08 Challenges of Deploying AI on Edge Devices

13:06 Techniques for Efficient Edge Model Training

15:44 The State of AI Hype and Reality

19:19 Evaluating AI Models and Benchmarks

24:09 Future of AI Architectures Beyond Transformers

31:05 Innovations in Model Architecture

36:28 The Importance of Iteration in AI Development

39:24 Understanding State Space Models

42:53 Advice for Aspiring Machine Learning Professionals

48:53 The Quest for Quality Data

52:56 Integrating User Feedback into AI Systems

58:13 Lessons from Machine Learning for Life

On this episode of Learning from Machine Learning, I had the privilege of speaking with Aman Khan, Head of Product at Arize AI. Aman shared how evaluating AI systems isn't just a step in the process—it's a machine learning challenge in of itself. Drawing powerful analogies between mechanical engineering and AI, he explained, "Instead of tolerances in manufacturing, you're designing for non-determinism," reminding us that complexity often breeds opportunity.

Aman's journey from self-driving cars to ML evaluation tools highlights the critical importance of robust systems that can handle failure. He encourages teams to clearly define outcomes, break down complex systems, and build evaluations into every step of the development pipeline.

Most importantly, Aman's insights remind us that machine learning—much like life—is less deterministic and more probabilistic, encouraging us to question how we deal with the uncertainty in our own lives.

Thank you for listening. Be sure to subscribe and share with a friend or colleague . Until next time... keep on learning.

Available on Youtube: https://youtu.be/v0eTTn7ZPEc

Available on Substack: https://mindfulmachines.substack.com/p/aman-khan-arize-evaluating-ai-designing?r=eykwy

In this episode of Learning from Machine Learning, we explore the intersection of pure mathematics and modern data science with Leland McInnes, the mind behind an ecosystem of tools for unsupervised learning including UMAP, HDBSCAN, PyNN Descent and DataMapPlot. As a researcher at the Tutte Institute for Mathematics and Computing, Leland has fundamentally shaped how we approach and understand complex data.

Leland views data through a unique geometric lens, drawing from his background in algebraic topology to uncover hidden patterns and relationships within complex datasets. This perspective led to the creation of UMAP, a breakthrough in dimensionality reduction that preserves both local and global data structure to allow for incredible visualizations and clustering. Similarly, his clustering algorithm HDBSCAN tackles the messy reality of real-world data, handling varying densities and noise with remarkable effectiveness.

But perhaps what's most striking about Leland isn't just his technical achievements – it's his philosophy toward algorithm development. He champions the concept of "decomposing black box algorithms," advocating for transparency and understanding over blind implementation. By breaking down complex algorithms into their fundamental components, Leland argues, we gain the power to adapt and innovate rather than simply consume.

For those entering the field, Leland offers poignant advice: resist the urge to chase the hype. Instead, find your unique angle, even if it seems unconventional. His own journey – applying concepts from algebraic topology and fuzzy simplicial sets to data science – demonstrates how breakthrough innovations often emerge from unexpected connections.

Throughout our conversation, Leland's passion for knowledge and commitment to understanding shine through. His approach reminds us that the most powerful advances in data science often come not from following the crowd, but from diving deep into fundamentals and drawing connections across disciplines.

There's immense value in understanding the tools you use, questioning established approaches, and bringing your unique perspective to the field. As Leland shows us, sometimes the most significant breakthroughs come from seeing familiar problems through a new lens.

Resources for Leland McInnes

Leland’s Github

• UMAP
• HDBSCAN
• PyNN Descent
• DataMapPlot
• EVoC

References

• Maarten Grootendorst
  • Learning from Machine Learning Episode 1
• Vincent Warmerdam - Calmcode
  • Learning from Machine Learning Episode 2
• Matt Rocklin
• Emily Riehl - Category Theory in Context
• Lorena Barba
• David Spivak - Fuzzy Simplicial Sets
• Improving Mapper’s Robustness by Varying Resolution According to Lens-Space Density

Learning from Machine Learning

• Youtube
• https://mindfulmachines.substack.com/