#+title: My learning on Neural network
* Take it with grain of salt
- I am not a AI engineer
- I just want to understand, how neural network work and how it is used ?
* Question
** Machine Learning terms
- ML Algorithms
Certainly! Here are some common machine learning algorithms categorized by their types:
*** Alogrithms
1. **Supervised Learning Algorithms**:
- **Linear Regression**: Used for predicting a continuous value
based on input features.
- **Logistic Regression**: Used for binary classification tasks.
- **Support Vector Machines (SVM)**: Effective for both
classification and regression tasks by finding the optimal
hyperplane that separates classes.
- **Decision Trees**: Tree-like models where each internal node
represents a feature, each branch represents a decision rule, and
each leaf node represents the outcome.
- **Random Forest**: Ensemble learning method that builds multiple
decision trees and merges them to improve accuracy.
- **Gradient Boosting Machines (GBM)**: Builds trees sequentially,
where each tree corrects the errors of the previous one.
- **Nearest Neighbors**: Predicts the target value for an
observation by using the target values of the k-nearest
neighbors.
2. **Unsupervised Learning Algorithms**:
- **K-Means Clustering**: Divides data into clusters based on
similarity.
- **Hierarchical Clustering**: Builds a tree of clusters where each
node is a cluster consisting of the nodes below it.
*** TLDR
- Regression
- How one thing affects another, Like eat more candy, you might get tummy ache.
- Classification
- Classify whether particular image contain cat or dog
- Clustering
- Take my relevant phots
- NLP (Natural Language Processing)
** What is Neural network ?
- Lecture 1 slides: MIT intro to deep learning
** How neural network learn
- Training
- Lecture 1 slides
- Backpropagation
- Andrej karpathy create neural network from scratch
- https://www.youtube.com/playlist?list=PLAqhIrjkxbuWI23v9cThsA9GvCAUhRvKZ
** Running a neural network
- Inference
- Andrej karpathy LLM video
- https://www.youtube.com/watch?v=zjkBMFhNj_g
** Different learning
- Neural network understand only numbers, right ?
*** Supervised
- Convolutional Neural Networks (CNNs): Particularly effective for image recognition tasks.
- Recurrent Neural Networks (RNNs): Suitable for sequential data like text and speech.
- Long Short-Term Memory (LSTM): A type of RNN that can learn long-term dependencies.
*** Unsupervised
- Generative models
- GAN
- Cycle GAN
- One domain to another domain
- Deepfakes, audio & video
*** Reinforncement Learning
- Rewards
- Q-Learning: A model-free reinforcement learning algorithm that learns a policy telling an agent what action to take under what circumstances.
- Deep Q-Networks (DQN): Combines Q-learning with deep neural networks to handle high-dimensional state spaces.
- Policy Gradient Methods: Directly optimize the policy function that maps states to actions.
- Alpha go
** Everything comes with price, What is price for AI ?
- Nothing is 100% perfect, No silver bullet
- Energy
- Computation
- E-waste
- Related article
- https://www.goldmansachs.com/intelligence/pages/gen-ai-too-much-spend-too-little-benefit.html
- https://www.goldmansachs.com/intelligence/topics/artificial-intelligence.html
** Learnings from ML/AI Developers
- Saama technologies, Chennai
- Samma connect meetup
- Sometimes normal machine learning model can solve a problem with better accuracy.
** Is NN can learn anything ?
- Lecture 6 slides: MIT intro to deep learning
* Resources
- ML Roadmap
- https://roadmap.sh/ai-data-scientist
- Deep learning: Alex & Ava MIT 2024 course
- http://introtodeeplearning.com/
- Other MIT course
- Stanford CS25: Transformers United
- https://www.youtube.com/playlist?list=PLoROMvodv4rNiJRchCzutFw5ItR_Z27CM
- Standford Seminar: ML Explainability
- https://www.youtube.com/playlist?list=PLoROMvodv4rPh6wa6PGcHH6vMG9sEIPxL
* Audio model
- https://github.com/Picovoice/picovoice
- https://github.com/rhasspy/piper
- https://github.com/mozilla/TTS
1 Like