deep-learning
Deep Learning expertise from Google DeepMind - creators of AlphaGo, Transformers, and BERT. Master neural network architectures, training at scale, and production ML. Use when building ML models, trai
SKILL.md.claude/skills/deep-learning/SKILL.md
GitHubDeep Learning - Google DeepMind Expertise
Expert: Dr. Emily Rodriguez (Google DeepMind, 10 years) Level: 10/10 - Google DeepMind researcher
Overview
Deep Learning at Google scale - neural networks powering billions of users across Search, YouTube, Photos, Assistant. Not toy models - production ML at the highest scale.
When to Use What
Computer Vision:
- Image classification → CNNs - Object detection → Faster R-CNN, YOLO
- Vision Transformers → ViT
Natural Language:
- Text understanding → Transformers - BERT, T5, PaLM models
- Fine-tuning for tasks
Sequence Modeling:
- Time series → RNNs, LSTM - Audio/speech → Wav2Vec, Conformer
Core Workflow
- Data preparation - Clean, augment, split datasets
- Model selection - Choose architecture for task
- Training - Use TPUs, distributed training 4. Evaluation - Rigorous metrics, validation
- Optimization - Hyperparameter tuning, architecture search
- Deployment - TensorFlow Serving, optimization 7. Monitoring - Track performance, detect drift
Google's Innovations
Transformers (2017): Attention mechanism revolutionized NLP BERT (2018): Bidirectional language understanding Vision Transformer (2020): Transformers for images PaLM (2022): 540B parameter language model Gemini (2023): Multimodal AI
Training at Scale
Google Infrastructure:
- TPU v4/v5 pods (thousands of chips)
- Distributed training strategies
- Mixed precision (bfloat16)
- Gradient accumulation
Production ML at Google
Powers:
- Google Search: Billions of queries/day
- YouTube: Video recommendations, 1B hours/day
- Google Photos: 4 trillion photos, auto-tagging
- Google Translate: 100+ languages
- Google Assistant: Natural conversations
Best Practices
- Start with pre-trained models - Transfer learning
- Use TPUs for large models - Cost-effective training
- Monitor training closely - TensorBoard, W&B
- Validate rigorously - Multiple metrics, holdout sets
- A/B test in production - Measure real impact
Related Skills
- gcp-expert - Vertex AI, TPUs
- nlp-expert - NLP applications
- ml-infrastructure - MLOps
Last Updated: 2026-02-03 Expert: Dr. Emily Rodriguez (Google DeepMind, 10 years) - AlphaGo, Transformers