Probability and Statistics for Deep Learning

Probability and Statistics for Deep Learning

Deep Learning is often called “Statistical Learning” and approached by many experts as statistical theory of the problem to find the best model approximation from a given collection of data. These models are produced today by Generative Deep Learning, which is a key to unlocking the latest sophisticated form of AI, comparable to human’s form of intelligence. AI is already present in our everyday lives, for example face recognition at airports, background generation in video conferencing, etc. You cannot understand CHAT GPT and Large LLMs if you don’t understand generative modeling and deep learning.
This course will introduce important concepts of probability theory and statistics which are the foundation of today’s Deep Learning. It will cover many important algorithms and modelling used in discriminative and generative deep learning models. In addition, the course will introduce tools and underlying mathematical concepts of data interpretation that work with specific models of deep neural networks. Upon completion of this course, you will have acquired the background in probability and statistics necessary for Machine Learning, and have the ability to use TensorFlow to create and train neural networks for specific practical problems.
 
Course Highlights:

• Knowledge of probability theory basic rules: conditional probability, joint probabilities, marginal probability
• Quantify events: Random Variables and becoming familiar with representing big data
• Important Probabilistic Distributions for Deep Learning
• Important Statistical Learning concepts: Entropy, KL divergence
• Statistical Estimators: MLE and MAP
• Logistic Regression and probabilistic models for classification (Kernels)
• Bayesian concept learning
• Discriminative models:
  • Supervised Learning Algorithms: Decision Trees and Random Forest
  • Unsupervised Learning Algorithms: K-Means & Gaussian models (mixture model)
• Structured Probabilistic Models: Probabilistic Graphical Model, Markov Model
• Generative modeling: what are the properties of good generative model,  (transformers, LLMs, Multimodal)
• Utilize TensorFlow 2.9 or PyTorch for real world example

Software: TensorFlow 2.9 Probability, PyTorch, Python 3.9.
We will use tools which are opensource and you will run it on UCSD Extension server using virtual environment setup for you with Jupyter notebook. 
 
Course typically offered: Online during our Spring and Fall academic quarters.

Prerequisites: Basic knowledge of Linear Algebra - concept of vectors and matrices.

Next steps: Upon completion, consider additional coursework in our specialized certificate in Machine Learning Methods to continue learning.

More information: For more information about this course, please contact unex-techdata@ucsd.edu.