Published Feb 8, 2025

Want to Understand Neural Networks? Think Elastic Origami! - Prof. Randall Balestriero

Explore the intricate geometry of neural networks with Professor Randall Balestriero as he unveils how spline theory and geometric insights can enhance understanding and robustness, tackle learning challenges like 'grokking,' and address advanced techniques for safeguarding large language models.

Episode Highlights

Topics covered

Questions from this episode

- How do you leverage different models in machine learning?
  Asked by 24 people
- What techniques are used with large language models (LLMs)?
  Asked by 23 people
- Tell me something unique about large language models
  Asked by 20 people
- Why are neural networks hard to explain?
  Asked by 13 people
- How do vector embeddings work in the context of the episode Mindscape 280 | François Chollet on Deep Learning and the Meaning of Intelligence and the clip Understanding Language Models?
  Asked by 12 people
- What challenges arise in data sets?
  Asked by 8 people
- How do we explain neural networks?
  Asked by 5 people
- How can I leverage pretrained models?
  Asked by 5 people
- Teach me about neural networks
  Asked by 5 people
- How do vector embeddings work in the context of the episode Mindscape 280 | François Chollet on Deep Learning and the Meaning of Intelligence and the clip Understanding Language Models?
  Asked by 4 people
- What's next in large language models (LLMs) as discussed in the episode Richard Socher: Re-Imagining Search and the clip Pretraining in Natural Language Processing?
  Asked by 2 people
- Neural networks explained
  Asked by 2 people
- What is the concept of emergence?
  Asked by 1 person
- What is the best insight on prompt engineering and engaging large language models (LLMs) from the episode Unpacking 3 Types of Feature Stores // Simba Khadder // #265 and the clip Evaluating Model Performance?
  Asked by 1 person
- I want to understand what a "Neural Network" is
  Asked by 1 person

Episode Highlights

Related Episodes

061: Interpolation, Extrapolation and Linearisation (Prof. Yann LeCun, Dr. Randall Balestriero)
Answers 383 questions
#86 - Prof. YANN LECUN and Dr. RANDALL BALESTRIERO - SSL, Data Augmentation, Reward isn't enough [NEURIPS2022]
Answers 383 questions
#041 - Biologically Plausible Neural Networks - Dr. Simon Stringer
Answers 383 questions
#60 Geometric Deep Learning Blueprint (Special Edition)
Answers 383 questions
#71 - ZAK JOST (Graph Neural Networks + Geometric DL) [UNPLUGGED]
Answers 383 questions
#063 - Prof. YOSHUA BENGIO - GFlowNets, Consciousness & Causality
Answers 383 questions
Understanding Deep Learning - Prof. SIMON PRINCE [STAFF FAVOURITE]
Answers 383 questions
Unlocking the Brain's Mysteries: Chris Eliasmith on Spiking Neural Networks and the Future of Human-Machine Interaction
Answers 383 questions
Neel Nanda - Mechanistic Interpretability (Sparse Autoencoders)
Answers 383 questions
#69 DR. THOMAS LUX - Interpolation of Sparse High-Dimensional Data
Answers 383 questions
Prof. Chris Bishop's NEW Deep Learning Textbook!
Answers 383 questions
Dr. Paul Lessard - Categorical/Structured Deep Learning
Answers 383 questions
Sepp Hochreiter - LSTM: The Comeback Story?
Answers 383 questions
Ben Goertzel on "Superintelligence"
Answers 383 questions
#040 - Adversarial Examples (Dr. Nicholas Carlini, Dr. Wieland Brendel, Florian Tramèr)
Answers 383 questions

Want to Understand Neural Networks? Think Elastic Origami! - Prof. Randall Balestriero

Topics covered

Popular Clips

Questions from this episode

Episode Highlights

Neural Geometry

Learning ChallengesRandall Balestriero explores the intricate dynamics of neural network training, emphasizing the phenomenon of "grokking" and its implications for robustness. He also delves into the role of intrinsic dimensionality and feature biases in shaping network behavior and performance.

Learning Challenges

RLHF and LLMs

Related Episodes