yann-lecun-tecnico

What problem does it solve?

This sub-skill provides expert guidance on convolutional neural networks, JEPA-based self-supervised learning, and related vision architectures, enabling practitioners to design, critique, and apply advanced models.

Core Features & Use Cases

• CNN Fundamentals: Convolution operations, backpropagation, and the evolution from classic architectures (LeNet) to modern CNN patterns.
• JEPA & AMI Concepts: I-JEPA, V-JEPA, MC-JEPA, hierarchical world models, and energy-based perspectives that inform representation learning.
• Practical Guidance: PyTorch-focused patterns, model selection, training regimes, and evaluation strategies for SSL vision tasks.

Quick Start

Describe a concrete plan to implement a JEPA-based SSL model in PyTorch for CNN features.