# Datawhale Course
**Course:** AI Research Foundations
**Compiled by:** Datawhale
---
## Learn Artificial Intelligence from Top Experts
Google DeepMind, in collaboration with **University College London (UCL)**, has launched a free course — **AI Research Foundations**.
This course is designed to help learners:
- Gain a **deep understanding of Transformers**.
- Develop **hands-on experience** in building and fine-tuning modern language models.
It is now **fully available** on the Google Skills platform.
🔗 **Course Homepage:** [https://www.skills.google/collections/deepmind](https://www.skills.google/collections/deepmind)
---
## Guided by the Head of Gemini
**Oriol Vinyals**:
- Vice President of Research & Head of Deep Learning at Google DeepMind
- Co-Lead for Gemini
- Co-inventor of the *seq2seq* model
- Key contributor to TensorFlow development
- Led the **AlphaStar** project that defeated top StarCraft II players
Recognized as a **leading authority** in both **deep learning** and **reinforcement learning**.
---
## Course Overview
### **Course 01: Build a Small Language Model Step-by-Step**
Focus: Fundamentals of language models + complete ML development workflow.
**You will learn:**
1. **Language Model Fundamentals**
- Predicting the next word
- Probability in modeling language patterns
2. **Traditional Methods & Limitations: N-gram Models**
- How N-gram models work
- Limitations in long-range dependencies
3. **Modern Architectures: Transformers**
- Core architecture powering today’s LLMs
- Comparison with N-grams
4. **Model Training & Evaluation**
- From data preparation to training a Small Language Model (SLM)
- Techniques for evaluating performance
---
### **Course 02: Representing Your Language Data**
Focus: Converting text into machine-readable formats with **Tokenization** and **Embeddings**.
**You will learn:**
1. **Data Processing Workflow** — Preprocess raw text for use in language models.
2. **Tokenization Techniques** — Characters, words, subwords; build a custom BPE tokenizer.
3. **Representing Meaning** — Map text into vectors/matrices that capture meaning and relationships.
4. **Ethics & Bias** — Detect dataset biases and apply the *data cards* process for ethical AI datasets.
---
💡 **Bonus for AI Content Creators**
Platforms like [AiToEarn官网](https://aitoearn.ai/) offer AI-powered content monetization tools — integrating generation, cross-platform publishing, analytics, and model ranking. Publish and earn across **Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X** efficiently.
---
### **Course 03: Designing and Training Neural Networks**
Focus: How to design, implement, and optimize neural networks.
**You will learn:**
1. **Training Mechanisms** — Backpropagation principles and how neural networks learn.
2. **Practical Implementation** — Code and evaluate a multilayer perceptron (MLP).
3. **Optimization & Troubleshooting** — Handle **overfitting** and **underfitting** effectively.
4. **Risk Awareness** — Consider potential risks, security concerns, and societal impacts.
---
### **Course 04: Exploring the Transformer Architecture**
Focus: The heart of modern LLMs — the **Transformer**.
**You will learn:**
1. **Transformer Mechanisms** — How models process prompts and predict next tokens.
2. **Attention Mechanism** — Visualize attention weights; understand masked & multi-head attention.
3. **Neural Network Techniques** — Additional concepts to improve NLP model performance.
---
⚡ **Tip for Advanced Learners**
Combining expertise in **neural network training** and **Transformer architecture** paves the way for building advanced AI systems. Monetizing these skills is possible via tools like [AiToEarn官网](https://aitoearn.ai/) — a global, open-source AI content creation and distribution platform.
---
