Aleksa Gordić - The AI Epiphany Courses

POET - Paired Open-Ended Trailblazer - Paper Explained

A paper walkthrough of POET, an evolutionary system that coevolves increasingly difficult environments and agents to create an open-ended learning curriculum.

• YouTube
• 45 minutes
• Self-Paced
• Free Video

Tensor Programs - Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer

Learn how μTransfer tunes hyperparameters on small neural networks and transfers them to larger models, reducing expensive search while preserving effective settings.

• YouTube
• 52 minutes
• Self-Paced
• Free Video

VOS - Learning What You Don't Know By Virtual Outlier Synthesis

A paper walkthrough of virtual outlier synthesis for detecting out-of-distribution inputs in image classification and object detection.

• YouTube
• 32 minutes
• Self-Paced
• Free Video

ConvNeXt- A ConvNet for the 2020s - Paper Explained

An explanation of ConvNeXt’s modernization of ResNets, showing how convolutional priors remain competitive with vision transformers in modern computer vision.

• YouTube
• 40 minutes
• Self-Paced
• Free Video

Machine Learning with Flax - From Zero to Hero

Learn Flax, a JAX-based neural-network library, by building models with Optax, custom modules, state handling, regularization, and a CNN on MNIST.

• YouTube
• 1 hour 18 minutes
• Self-Paced
• Free Video

A Neural Network Solves and Generates Mathematics Problems by Program Synthesis - Paper Explained

An explanation of how OpenAI Codex uses prompt-modified math problems and generated Python code to solve and generate university-level mathematics problems.

• YouTube
• 33 minutes
• Self-Paced
• Free Video

OpenAI GLIDE - Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models

A technical walkthrough of GLIDE's diffusion-and-transformer pipeline for text-guided photorealistic image generation, inpainting, guidance, and failure cases.

• YouTube
• 53 minutes
• Self-Paced
• Free Video

Efficient Geometry-Aware 3D Generative Adversarial Networks - GAN Paper Explained

This paper walkthrough explains EG3D's tri-plane scene representation, NeRF foundations, pose-conditioned generation, dual discrimination, and multi-view-consistent 3D GAN results.

• YouTube
• 43 minutes
• Self-Paced
• Free Video

Neural Descriptor Fields- SE(3)-Equivariant Object Representations for Manipulation Paper Explained

Explains SE(3)-equivariant neural descriptor fields for robust object grasping and placement across novel shapes and poses.

• YouTube
• 33 minutes
• Self-Paced
• Free Video

Coding a Neural Network from Scratch in Pure JAX - Machine Learning with JAX - Tutorial 3

Build and train an MLP classifier on MNIST in pure JAX, then visualize its weights and embeddings and inspect dead neurons.

• YouTube
• 1 hour 26 minutes
• Self-Paced
• Free Video

Machine Learning with JAX - From Hero to HeroPro+

Build complex neural networks in JAX with PyTrees, automatic differentiation, and multi-device training.

• YouTube
• 1 hour 9 minutes
• Self-Paced
• Free Video

The Sensory Neuron as a Transformer - Permutation-Invariant Neural Networks for Reinforcement Learning - Paper Explained

A visual paper explanation of attention-based agents that preserve reinforcement-learning performance when image observations are arbitrarily permuted.

• YouTube
• 33 minutes
• Self-Paced
• Free Video

Neural Search with Jina AI - Open-Source ML Tool Explained

Explore Jina AI’s neural search framework through a Fashion-MNIST image-search example, covering indexing, embeddings, and result visualization.

• YouTube
• 32 minutes
• Self-Paced
• Free Video

Do Vision Transformers See Like Convolutional Neural Networks - Paper Explained

Explains how vision transformers and convolutional neural networks learn different representations through receptive fields, data scale, skip connections, and token flow.

• YouTube
• 35 minutes
• Self-Paced
• Free Video

Emerging Properties in Self-Supervised Vision Transformers - Paper Explained

Explains DINO, a self-supervised method for vision transformers, through its teacher network, multi-crop strategy, attention maps, emergent segmentation, and collapse analysis.

• YouTube
• 32 minutes
• Self-Paced
• Free Video