C ourse Syllabus

Machine Learning Injection

Philosophy

Theory & practice Fast paced- Co mm i - iv

un ty dr en
The Deep Atlas program is designed with a The Deep Atlas curriculum is a tour de force The Deep Atlas community is comprised of a
careful balance of fundamental theory and of knowledge compression: optimally select group of talented technologists who
applied problem-solving techniques. sequenced for maximum time efficiency. want to advance their careers in AI/ML.

Technologies

Schedule

(A) Full-Time Immersive (2 Weeks) (B) Nights & Weekends (4 Weeks) l

C ass Structure
Mo - ri 7:30am - 6:00pm
n F (PT ) F - 7:30am - 3:30pm
ri Sun (PT ) 1. In both program options, please reserve extra hours
outside of class hours to complete assignments.

Br k 10:30am - 11:30am
ea (PT ) Br k 10:30am - 11:30am
ea (PT ) 2. ach full day of class consists of pair programming
E

Br k 3:30pm - 4:30pm
ea (PT ) Mo - W 4:30pm - 6:00pm
n ed (PT )
sessions, live lectures, group discussions, and more.

Ov rvi we e

lifecycle of models from

inception to production

Coursework Focus

Advanced Material
Prompt Engineering AI Engineering ML Engineering ML Operations

systematically crafting inputs implementing pre-trained training, deploying, and

ML Research
to guide a language model models for real-world use evaluating models

advancing the field through

architectural innovations

Curriculum

F undamentals P recourse

Problem Framing Data Tools & Techniques Probability & Statistics

Begin the journey of building an intuition for the kinds G ain experience with NumPy, Matplotlib, and Pandas to Develop an intuition for probability, distributions, and
of problems that are well-suited for machine learning. clean, process, manipulate and visualize data. convolutions to reason about model functioning.

E nvironment Setup S imple Linear Regression I ntro to Neural Networks

Establish the developer tools, workflows, and system Train a model to predict house prices using gradient Build a foundational understanding of gradient
configurations to hit the ground running. descent, visualizing the cost curve and descent path. calculation and backpropagation algorithms.

L inear Algebra Fundamentals B uilding with LLMs The ML Terrain

Understand how vectors and matrices represent data U se LangChain to chain prompts and create an agent Explore the diverse, rapidly evolving ML landscape,
and transformations in high-dimensional spaces. that interacts with third-party sources and APIs. considering job roles, breakthroughs, and ethics.

Shallow Learning

Sp atializing Data M ultiple Linear Regression C lassification & Clustering

Visualize matrix transformations, dimensionality D evelop a prediction model using Weights and Biases, A pply K-Nearest Neighbors, Support Vector Machines
reduction/expansion, and semantic meaning. S cikit-learn, and Matplotlib to predict student grades. (SVM ) Ls , ogistic Regression, and K-Means clustering.

V ector & Semantic Search G radient Boosted Trees N ondeterministic Paradigm Shift
Explore multidimensional data by implementing vector Train a gradient boosted model with XGBoost and tune Contrast the deterministic paradigm of software
databases, embedding models, and similarity search. hyperparameters to predict home sale prices. engineering with the nondeterministic paradigm of ML.

I nference & Training Mechanics Taxonomy of Shallow Algorithms A nomaly Detection

Explore the relationship between data features, Explore shallow algorithms and their systematic Use the IsolationForest algorithm to perform anomaly
parameters, and hyperparameters in the training cycle. applicability across diverse machine learning tasks. detection on bank transactions to detect fraud.

M odel Development Lifecycle Dimensionality Reduction M odule Project

Tour the iterative training process from benchmarking U se Scikit-learn to apply PCA, t-SNE, and UMAP on Solve a real-world classification problem by preparing
and measurement to optimization and validation. MNIST data and visualize the projections. data, engineering features, and optimizing a model.

Deep Learning

N eural Network Optimization Algorithms L ayer-Level Architecture Sy stem-Level Architectures

Explore gradient descent optimization algorithms and Explore the anatomical structure of layer types like Explore multi-network architectures like generative
their role in efficiently training neural networks. convolutional, dropout, recurrent, pooling, and more. adversarial networks and encoder-decoder networks.

A natomy of a Neural Network E nhancing a Neural Network A ttention & Transformers

Explore neural networks from the bottom-up: input Experiment with manipulating the layers of a neural plore the transformer architecture and attention
Ex

features, weights, activations, embeddings, and layers. network to enhance accuracy and extend functionality. mechanism, which enable context-aware predictions.

Deep Neural Networks Taxonomy of Neural Networks I mplementing Attention

nstantiate and train a neural network using PyTorch to
I Explore neural network architectures and their Build an attention head and explore weighted dot
optically identify handwritten digits using MNIST data. systematic applicability across diverse task types. products, the key components of a transformer model.

Debugging & Testing The ML Research Workflow M odule Project

Learn to validate data, debug and optimize models, and Explore the iterative process from problem framing and Pick a time series dataset and build an LSTM-based
monitor performance metrics across an ML pipeline. data cleaning to convergence and productionization. neural network using PyTorch to forecast values.

v
Le eraged L M
L everaged ML Strategies LLM Overview ML Ops Fundamentals
Explore the taxonomy of transfer learning techniques Ex plore the capabilities, limitations, and frontiers of Explore key principles, tools, and procedures for
and their respective use cases (e.g., PEFT with LoRA). large language models from agents to RLHF, and more. improving automation, reproducibility, and monitoring.

Transfer Learning with Embeddings R etrieval Augmented Generation Deploying Models to AWS
C ompare the performance of training an image Vectorize a data corpus into a vector database and run U se TorchServe to package and deploy an image
classifier directly on image data vs. using embeddings. a semantic search before inference to enhance output. recognition model for remote, server-based inference.

Transfer Learning with Fine-Tuning The Rules of Machine Learning M odule Project
F 18
ine-tune Resnet to perform a binary classification Explore best practices for reproducibility, performance, Select a transfer learning, RAG, or LLM API-driven
task with more accuracy compared to the base version. data quality, and iterative improvement. strategy to solve a real problem from end-to-end.

deepatlas.ai Curriculum and schedule subject to change