Курсы базовой части
-
Math for AI
calculus:
function and function limit
logarithms
derivative and partial derivative
gradient and chain rule
convex function
definite and indefinite integrals
Taylor series
probability theory:
discrete and continuous random variables
probability function / density function
independent events
conditional probability
Bayes’ rule
expectation, variance
joint probability
covariance
splines
PCA (Principal component analysis)
entropy, cross-entropy, relative entropy
algebra:
vectors
matrices
operation with vectors and matrices
determinant
eigenvalues and eigenvector
spectral decomposition
SVD (Singular Value Decomposition) -
Python for AI
basics recap:
git flow
project structure
code styling
documentation standards
decorators
data manipulation:
efficient math operations (NumPy)
accelerating numerical computations (Numba)
exploratory data analysis (pandas)
performance speed optimization (Cython)
neural networks (using PyTorch):
tensor operations
autograd and computation graphs
neural network design (torch.nn, layers, forward propagation)
training and optimization (loss functions, optimizers like SGD and Adam, training loops)
data handling (Datasets and DataLoader classes)
data visualization:
visualization fundamentals (Matplotlib)
high-level plotting (Seaborn)
interactive visualizations (Plotly)
web-oriented tools (Bokeh) -
AI: from Basics to Transformers. Part 1
basics:
supervised learning
regression
regularization
classification
metrics
logistic regression
Курсы продвинутой части
-
MLOps
data:
data management (ClickHouse)
data processing manipulations (Plotly Dash, Streamlit)
model:
model engineering: hyperparameters tuning (Optuna)
prepare model for production (TorchScript, ONNX)
ML pipeline:
training/validation/evaluation logging
experiments tracking (Neptune.ai/MLflow)
serving:
deployment tools (Docker, FastAPI) -
Decision making in AI
project structure
task formulation
task analysis
data collection
data annotation
exploratory data analysis
proof of concept (PoC) development
analysis of PoC results
PoC improvement
getting to the final result
presenting results -
AI: from Basics to Transformers. Part 2
intro to DL:
intro to deep learning
multilayer perceptron
backpropagation
deep learning mindset
CNNs:
convolutional neural nets
building convolutional architectures
residual networks
CNN training best practices
CV:
image retrieval (NEW)
image segmentation (NEW)
transformers:
attention mechanism in transformer
multi-head attention
properties of MHA
transformer encoder
transformer decoder
model interpretation
transformer training best practices
vision transformer
improving ViT
time series forecasting with transformer (NEW) -
AI: from Transformers to LLM. Part 3
introduction to language models
large language models
RoPE (NEW)
GQA (NEW)
SwiGLU (NEW)