Yannic Kilcher Courses

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

Big Self-Supervised Models Are Strong Semi-Supervised Learners

Explore SimCLRv2, a powerful semi-supervised learning approach combining self-supervised pre-training, fine-tuning, and distillation to achieve state-of-the-art results with minimal labeled data.

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

On the Measure of Intelligence by François Chollet - Human Priors - Paper Explained

Explores the relationship between intelligence, generality, and prior knowledge, focusing on human priors as a basis for comparing human and AI intelligence. Discusses optimizing for generality and the role of experience in skill acquisition.

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

Image GPT- Generative Pretraining from Pixels - Paper Explained

Exploring generative pretraining for image processing, adapting NLP techniques to visual tasks. Discusses model architecture, experiments, and potential applications in computer vision.

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

Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning - Paper Explained

Explore BYOL, a novel self-supervised learning approach that outperforms baselines without negative samples. Learn about its innovative use of online and target networks for image representation.

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

SynFlow - Pruning Neural Networks Without Any Data by Iteratively Conserving Synaptic Flow

Explores SynFlow, a novel data-agnostic algorithm for pruning neural networks at initialization, avoiding layer collapse and achieving maximum compression capacity without training or data.

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

VirTex- Learning Visual Representations from Textual Annotations

Explore VirTex, a novel approach to visual representation learning using high-quality image captions, outperforming traditional methods with fewer images for various computer vision tasks.

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

Linformer - Self-Attention with Linear Complexity

Detailed explanation of the Linformer model, which reduces self-attention complexity in transformers from O(n²) to O(n), improving efficiency while maintaining performance in NLP tasks.

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

TransCoder - Unsupervised Translation of Programming Languages

Explore unsupervised neural machine translation for code migration between Python, C++, and Java. Learn about shared embeddings, objectives, evaluation, and results of this innovative approach.

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

BLEURT - Learning Robust Metrics for Text Generation

Explora BLEURT, una métrica de evaluación basada en BERT para generación de texto, su preentrenamiento con datos sintéticos y su capacidad de modelar juicios humanos con pocos ejemplos.

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

Synthetic Petri Dish - A Novel Surrogate Model for Rapid Architecture Search

Explore a novel approach to Neural Architecture Search using a Synthetic Petri Dish model, which evaluates architectural motifs in small networks with synthetic data to accelerate performance prediction and optimization.

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

CornerNet- Detecting Objects as Paired Keypoints

Explores CornerNet, a novel object detection approach using paired keypoints. Covers corner pooling, heatmap and embedding outputs, loss functions, and experimental results.

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

Synthesizer - Rethinking Self-Attention in Transformer Models

Explores alternative attention mechanisms in Transformer models, proposing the Synthesizer model that learns synthetic attention weights without token-token interactions, showing competitive performance across various NLP tasks.

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

GPT-3 - Language Models Are Few-Shot Learners

Comprehensive analysis of GPT-3's capabilities, exploring its performance on various NLP tasks and discussing its potential impact on AI research and applications.

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