Yannic Kilcher Courses

Transformers Are RNNs- Fast Autoregressive Transformers With Linear Attention

Explore a novel approach to transformers using linear attention, reducing computational complexity and revealing connections to RNNs. Gain insights into faster, more efficient deep learning models for long sequences.

• YouTube
• 48 minutes
• On-Demand
• Free Video

BERTology Meets Biology - Interpreting Attention in Protein Language Models

Explores how BERT models trained on protein sequences learn biological properties, revealing insights into protein structure, binding sites, and biophysical characteristics through attention mechanism analysis.

• YouTube
• 37 minutes
• On-Demand
• Free Video

GShard- Scaling Giant Models with Conditional Computation and Automatic Sharding

Detailed explanation of Google's 600 billion parameter transformer for multilingual translation, focusing on scaling techniques, mixture-of-experts, and automatic sharding implementation on TPUs.

• YouTube
• 1 hour 13 minutes
• On-Demand
• Free Video

Object-Centric Learning with Slot Attention

Explanation of a novel deep learning module for object-centric visual scene understanding. Covers architecture, algorithm, experiments, and implications of Slot Attention for improved object recognition and reasoning.

• YouTube
• 43 minutes
• On-Demand
• Free Video

Set Distribution Networks - A Generative Model for Sets of Images

Explores a novel framework for generating and encoding sets of images, introducing Set Distribution Networks that can create new object/identity sets while preserving key attributes.

• YouTube
• 59 minutes
• On-Demand
• Free Video

Direct Feedback Alignment Scales to Modern Deep Learning Tasks and Architectures

Explores Direct Feedback Alignment as a biologically plausible alternative to backpropagation, demonstrating its effectiveness in training modern deep learning architectures for various challenging tasks.

• YouTube
• 35 minutes
• On-Demand
• Free Video

Discovering Symbolic Models from Deep Learning with Inductive Biases

Exploring how Graph Neural Networks and symbolic regression can derive accurate symbolic equations from observational data, combining deep learning with physics-inspired inductive biases.

• YouTube
• 46 minutes
• On-Demand
• Free Video

How I Read a Paper- Facebook's DETR

Explore the process of understanding Facebook AI's DETR paper through a step-by-step analysis, from title to conclusions, with insights on interpreting research effectively.

• YouTube
• 45 minutes
• On-Demand
• Free Video

RepNet - Counting Out Time - Class Agnostic Video Repetition Counting in the Wild

Explore RepNet's innovative approach to video repetition counting using temporal self-similarity matrices, synthetic data, and a constrained model for class-agnostic period prediction in real-world videos.

• YouTube
• 37 minutes
• On-Demand
• Free Video

SIREN - Implicit Neural Representations with Periodic Activation Functions

Explore SIREN neural networks for representing complex signals like images and 3D shapes. Learn about their unique periodic activation functions, initialization, and applications in solving differential equations.

• YouTube
• 56 minutes
• On-Demand
• Free Video

OpenAssistant - ChatGPT's Open Alternative

Explore OpenAssistant, an open-source alternative to ChatGPT. Learn about the project's goals, data collection process, and participate in various AI interaction tasks to improve the system.

• YouTube
• 36 minutes
• On-Demand
• Free Video

CICERO- An AI Agent That Negotiates, Persuades, and Cooperates With People

Explore Meta AI's Cicero, an advanced AI agent that excels in Diplomacy by integrating language models with strategic planning to negotiate and cooperate with humans effectively.

• YouTube
• 1 hour 1 minute
• On-Demand
• Free Video

The New AI Model Licenses Have a Legal Loophole - OpenRAIL-M of BLOOM, Stable Diffusion, etc.

Análisis crítico de las licencias de IA responsable, sus limitaciones y posibles consecuencias legales. Explora el dilema entre el uso abierto y las restricciones, proponiendo una alternativa llamada OpenRAIL++.

• YouTube
• 28 minutes
• On-Demand
• Free Video

ROME - Locating and Editing Factual Associations in GPT - Paper Explained & Author Interview

Explores mechanisms for storing and editing factual knowledge in GPT models, revealing key roles of mid-layer feed-forward modules and proposing targeted editing techniques with implications for model understanding and control.

• YouTube
• 1 hour 5 minutes
• On-Demand
• Free Video

Is Stability Turning into OpenAI?

Explore the controversy surrounding Stability AI's actions, including community takeovers, intellectual property disputes, and concerns about openness in AI development.

• YouTube
• 39 minutes
• On-Demand
• Free Video