All Roadmaps / Deep Learning Specialist

Deep Learning Interactive Graph Curriculum

Deep Learning Specialist

A deep-dive curriculum covering neural network calculus, PyTorch computational graphs, Computer Vision (CNNs, YOLO), NLP, Transformer attention, and Generative AI / RAG architectures.

5 Phases

13 Topics

~216h Total Time

Study Progress: 0 of 13 completed • 0h of ~216h 0%

Status:

Level:

🎯

Recommended Next Step Beginner ⏱️ 3 hours

Intelligent Agents & Problem Formulation

Graph Legend:

Recommended

Optional

Beginner

Intermediate

Advanced

Done • In Progress • Use J/K or click cards to explore

Curriculum Start

01 Phase 01 0/3 completed

Phase 1: Neural Networks from Scratch & PyTorch

Mathematical intuition of biological and artificial neurons, backpropagation calculus, and the PyTorch autograd engine.

beginner Recommended • 10 days

Perceptrons, Multi-Layer Perceptrons & Activations

Linear combinations, forward propagation, non-linear activation functions (Sigmoid, Tanh, ReLU, Leaky ReLU, GELU, Swish), and universal approximation theorem.

intermediate Recommended • 2 weeks

Backpropagation Calculus & Computational Graphs

Reverse-mode automatic differentiation, computing loss gradients with respect to weights and biases, and implementing a micro-framework from scratch.

intermediate Recommended • 2 weeks

PyTorch Optimization & Regularization

Modern optimization algorithms (Adam, AdamW, RMSProp), learning rate schedules (Cosine Annealing, OneCycleLR), Weight Decay, and Dropout.

02 Phase 02 0/3 completed

Phase 2: Computer Vision & Spatial Feature Learning

Spatial convolution kernels, image classification, transfer learning backbones, and real-time object detection.

beginner Recommended • 10 days

Computer Vision Basics & OpenCV Processing

Digital image representation, color channels (RGB, BGR, HSV), spatial filtering (Sobel, Gaussian blur, Canny edge detection), affine transformations, and thresholding.

intermediate Recommended • 2 weeks

Convolutional Neural Networks (CNNs) & Residual Networks

2D Convolution layers, padding (valid vs same), stride, pooling (MaxPool, AvgPool), receptive fields, and residual skip connections (ResNet, ConvNeXt).

advanced Recommended • 2 weeks

Object Detection & Real-Time Models (YOLO)

Anchor boxes, non-maximum suppression (NMS), intersection over union (IoU), two-stage (Faster R-CNN) vs one-stage detectors (YOLOv8, YOLOv11), and feature pyramid networks.

03 Phase 03 0/2 completed

Phase 3: Natural Language Processing & Sequence Modeling

Text preprocessing, word vector embeddings, recurrent networks, and sequence-to-sequence pipelines.

beginner Recommended • 10 days

Text Preprocessing, Tokenization & Vector Space Models

Regex normalization, stop words, lemmatization, TF-IDF, Word2Vec (Skip-Gram & CBOW), GloVe embeddings, and cosine semantic similarity.

intermediate Recommended • 2 weeks

Recurrent Neural Networks (RNNs, LSTMs, GRUs)

Hidden state recurrent loops, handling variable-length sequential data, vanishing gradient challenges, Long Short-Term Memory (LSTM) gates, and bidirectional processing.

04 Phase 04 0/2 completed

Phase 4: Transformer Architecture & Large Language Models

Self-attention mechanics, multi-head attention, positional encoding, encoder-decoder models, and foundational LLM architectures.

advanced Recommended • 3 weeks

Self-Attention & The Transformer Architecture

Scaled dot-product attention, Query, Key, and Value projections, Multi-Head Attention, Rotary Positional Embeddings (RoPE), LayerNorm, and RMSNorm.

advanced Recommended • 2 weeks

Pretrained LLM Ecosystem & Fine-Tuning (LoRA, QLoRA)

HuggingFace Transformers, model hubs, Parameter-Efficient Fine-Tuning (PEFT), Low-Rank Adaptation (LoRA), 4-bit quantization (bitsandbytes), and SFT pipelines.

05 Phase 05 0/3 completed

Phase 5: Generative AI, RAG & Agentic Systems

Production Retrieval-Augmented Generation, vector embeddings, LangChain / LangFlow architectures, and multi-agent coordination.

intermediate Recommended • 2 weeks

Retrieval-Augmented Generation (RAG) Architecture

Chunking strategies (recursive, semantic), vector embeddings, HNSW indexing, vector databases (Chroma, Pinecone, Qdrant), hybrid search, and re-ranking.

advanced Recommended • 2 weeks

Agentic AI & Autonomous Tool Calling (LangFlow & n8n)

ReAct reasoning loop (Reasoning + Acting), structured tool calling, multi-agent orchestration with LangGraph / n8n, memory persistence, and human-in-the-loop workflows.

advanced Optional • 2 weeks

Reinforcement Learning & Deep RL (PPO, Q-Learning)

Markov Decision Processes (MDP), value functions, Bellman equation, Deep Q-Networks (DQN), Policy Gradients, and Proximal Policy Optimization (PPO) used in RLHF.

Curriculum Breakdown

Phase 1: Neural Networks from Scratch & PyTorch

0/3 completed

Mathematical intuition of biological and artificial neurons, backpropagation calculus, and the PyTorch autograd engine.

Perceptrons, Multi-Layer Perceptrons & Activations

beginner ⏱️ 10 days

Linear combinations, forward propagation, non-linear activation functions (Sigmoid, Tanh, ReLU, Leaky ReLU, GELU, Swish), and universal approximation theorem.

Key Outcomes:

✓ Why non-linear activation functions are required for deep representation

✓ Vanishing and exploding gradient problems with Sigmoid/Tanh

✓ Softmax activation for multiclass probability calibration

Links: DeepLearning.AI Neural Networks and Deep Learning (Andrew Ng) Neural Networks and Deep Learning - Michael Nielsen

Backpropagation Calculus & Computational Graphs

intermediate ⏱️ 2 weeks

Reverse-mode automatic differentiation, computing loss gradients with respect to weights and biases, and implementing a micro-framework from scratch.

Key Outcomes:

✓ Analytical gradient computation across directed acyclic graphs (DAGs)

✓ Weight update step using Stochastic Gradient Descent (SGD) and Momentum

✓ Building an autograd scalar engine from first principles

Links: Andrej Karpathy: Building micrograd (Neural Networks: Zero to Hero) Backpropagation Calculus - 3Blue1Brown

PyTorch Optimization & Regularization

intermediate ⏱️ 2 weeks

Modern optimization algorithms (Adam, AdamW, RMSProp), learning rate schedules (Cosine Annealing, OneCycleLR), Weight Decay, and Dropout.

Key Outcomes:

✓ Why AdamW decouples L2 weight decay from gradient momentum

✓ Learning rate warm-up and cosine decay dynamics

✓ Using Mixed Precision (torch.cuda.amp / bfloat16) to speed up training

Links: Official PyTorch Tutorials & Recipes Deep Learning Crash Course for Beginners

Phase 2: Computer Vision & Spatial Feature Learning

0/3 completed

Spatial convolution kernels, image classification, transfer learning backbones, and real-time object detection.

Computer Vision Basics & OpenCV Processing

beginner ⏱️ 10 days

Digital image representation, color channels (RGB, BGR, HSV), spatial filtering (Sobel, Gaussian blur, Canny edge detection), affine transformations, and thresholding.

Key Outcomes:

✓ Manipulating image matrices and pixel data in OpenCV and NumPy

✓ Spatial convolution with 2D kernel filters

✓ Image augmentations (Random Crop, Flips, ColorJitter) using Albumentations

Links: OpenCV Course - Full Tutorial with Python OpenCV Official Bootcamp Course

Convolutional Neural Networks (CNNs) & Residual Networks

intermediate ⏱️ 2 weeks

2D Convolution layers, padding (valid vs same), stride, pooling (MaxPool, AvgPool), receptive fields, and residual skip connections (ResNet, ConvNeXt).

Key Outcomes:

✓ Spatial hierarchy extraction (edges -> textures -> parts -> objects)

✓ How residual skip connections solve gradient degradation in deep nets

✓ Global Average Pooling to eliminate dense FC parameters

Links: Stanford CS231n: CNNs for Visual Recognition Lectures Coursera: Convolutional Neural Networks (Andrew Ng)

Object Detection & Real-Time Models (YOLO)

advanced ⏱️ 2 weeks

Anchor boxes, non-maximum suppression (NMS), intersection over union (IoU), two-stage (Faster R-CNN) vs one-stage detectors (YOLOv8, YOLOv11), and feature pyramid networks.

Key Outcomes:

✓ Formulating bounding box regression + classification loss

✓ mAP@0.5 and mAP@0.5:0.95 detection benchmark evaluation

✓ Real-time edge inference with Ultralytics YOLO

Links: Ultralytics YOLO Documentation & Guides Computer Vision Essentials Course

Phase 3: Natural Language Processing & Sequence Modeling

0/2 completed

Text preprocessing, word vector embeddings, recurrent networks, and sequence-to-sequence pipelines.

Text Preprocessing, Tokenization & Vector Space Models

beginner ⏱️ 10 days

Regex normalization, stop words, lemmatization, TF-IDF, Word2Vec (Skip-Gram & CBOW), GloVe embeddings, and cosine semantic similarity.

Key Outcomes:

✓ Embedding words into continuous geometric vector spaces

✓ Cosine distance metrics for semantic similarity search

✓ Subword tokenization vs word-level out-of-vocabulary (OOV) bottlenecks

Links: How To Get Started with NLP - Towards Data Science TensorFlow's NLP Zero to Hero Playlist

Recurrent Neural Networks (RNNs, LSTMs, GRUs)

intermediate ⏱️ 2 weeks

Hidden state recurrent loops, handling variable-length sequential data, vanishing gradient challenges, Long Short-Term Memory (LSTM) gates, and bidirectional processing.

Key Outcomes:

✓ Forget, input, and output gating mechanisms in LSTM units

✓ Sequence-to-sequence encoder-decoder architectures with attention

✓ Why parallelization constraints led to the Transformer revolution

Links: Understanding LSTM Networks - Christopher Olah Natural Language Processing Pipeline Overview

Phase 4: Transformer Architecture & Large Language Models

0/2 completed

Self-attention mechanics, multi-head attention, positional encoding, encoder-decoder models, and foundational LLM architectures.

Self-Attention & The Transformer Architecture

advanced ⏱️ 3 weeks

Scaled dot-product attention, Query, Key, and Value projections, Multi-Head Attention, Rotary Positional Embeddings (RoPE), LayerNorm, and RMSNorm.

Key Outcomes:

✓ Deriving attention weights: softmax(QK^T / sqrt(d_k)) * V

✓ Computational complexity of full self-attention O(N^2)

✓ Decoder causal attention masking for autoregressive language generation

Links: The Illustrated Transformer by Jay Alammar Andrej Karpathy: Let's build GPT from scratch, in code

Pretrained LLM Ecosystem & Fine-Tuning (LoRA, QLoRA)

advanced ⏱️ 2 weeks

HuggingFace Transformers, model hubs, Parameter-Efficient Fine-Tuning (PEFT), Low-Rank Adaptation (LoRA), 4-bit quantization (bitsandbytes), and SFT pipelines.

Key Outcomes:

✓ Freezing base foundation weights and training rank-decomposition matrices

✓ 4-bit NormalFloat (NF4) quantization mechanisms in QLoRA

✓ Instruction fine-tuning datasets and formatting templates

Links: Hugging Face PEFT / LoRA Guide Generative AI for Developers Course

Phase 5: Generative AI, RAG & Agentic Systems

0/3 completed

Production Retrieval-Augmented Generation, vector embeddings, LangChain / LangFlow architectures, and multi-agent coordination.

Retrieval-Augmented Generation (RAG) Architecture

intermediate ⏱️ 2 weeks

Chunking strategies (recursive, semantic), vector embeddings, HNSW indexing, vector databases (Chroma, Pinecone, Qdrant), hybrid search, and re-ranking.

Key Outcomes:

✓ Building an end-to-end RAG question-answering pipeline

✓ Dense vector embeddings vs sparse BM25 keyword matching

✓ Cross-encoder re-ranking for high retrieval precision

Links: Learn RAG From Scratch (Full Course) Linux Foundation: RAG Introduction (RXM403)

Agentic AI & Autonomous Tool Calling (LangFlow & n8n)

advanced ⏱️ 2 weeks

ReAct reasoning loop (Reasoning + Acting), structured tool calling, multi-agent orchestration with LangGraph / n8n, memory persistence, and human-in-the-loop workflows.

Key Outcomes:

✓ Autonomous agent loops: thought -> plan -> tool call -> observe -> synthesize

✓ Building multi-agent visual workflows with LangFlow and n8n

✓ State graphs and cyclical graph execution with LangGraph

Links: AI Agents Fundamentals in 7 Minutes Building a Team of AI Agents in n8n (No-Code) Getting Started with LangFlow

Reinforcement Learning & Deep RL (PPO, Q-Learning)

advanced Optional ⏱️ 2 weeks

Markov Decision Processes (MDP), value functions, Bellman equation, Deep Q-Networks (DQN), Policy Gradients, and Proximal Policy Optimization (PPO) used in RLHF.

Key Outcomes:

✓ Exploration vs exploitation dilemma in dynamic environments

✓ Q-value approximation with deep neural networks

✓ PPO clipped surrogate objective for stable policy updates

Links: HuggingFace Deep Reinforcement Learning Course Reinforcement Learning By The Book Playlist

Beginner 45 min

Topic Title

Status:

What You Need to Know

Topic explanation description goes here.

Code Blueprint

Progress exported & copied to clipboard!

Deep Learning Specialist

Intelligent Agents & Problem Formulation

Phase 1: Neural Networks from Scratch & PyTorch

Perceptrons, Multi-Layer Perceptrons & Activations

Backpropagation Calculus & Computational Graphs

PyTorch Optimization & Regularization

Phase 2: Computer Vision & Spatial Feature Learning

Computer Vision Basics & OpenCV Processing

Convolutional Neural Networks (CNNs) & Residual Networks

Object Detection & Real-Time Models (YOLO)

Phase 3: Natural Language Processing & Sequence Modeling

Text Preprocessing, Tokenization & Vector Space Models

Recurrent Neural Networks (RNNs, LSTMs, GRUs)

Phase 4: Transformer Architecture & Large Language Models

Self-Attention & The Transformer Architecture

Pretrained LLM Ecosystem & Fine-Tuning (LoRA, QLoRA)

Phase 5: Generative AI, RAG & Agentic Systems

Retrieval-Augmented Generation (RAG) Architecture

Agentic AI & Autonomous Tool Calling (LangFlow & n8n)

Reinforcement Learning & Deep RL (PPO, Q-Learning)

Curriculum Breakdown

Phase 1: Neural Networks from Scratch & PyTorch

Perceptrons, Multi-Layer Perceptrons & Activations

Backpropagation Calculus & Computational Graphs

PyTorch Optimization & Regularization

Phase 2: Computer Vision & Spatial Feature Learning

Computer Vision Basics & OpenCV Processing

Convolutional Neural Networks (CNNs) & Residual Networks

Object Detection & Real-Time Models (YOLO)

Phase 3: Natural Language Processing & Sequence Modeling

Text Preprocessing, Tokenization & Vector Space Models

Recurrent Neural Networks (RNNs, LSTMs, GRUs)

Phase 4: Transformer Architecture & Large Language Models

Self-Attention & The Transformer Architecture

Pretrained LLM Ecosystem & Fine-Tuning (LoRA, QLoRA)

Phase 5: Generative AI, RAG & Agentic Systems

Retrieval-Augmented Generation (RAG) Architecture

Agentic AI & Autonomous Tool Calling (LangFlow & n8n)

Reinforcement Learning & Deep RL (PPO, Q-Learning)

No matching topics found

Topic Title

What You Need to Know

Code Blueprint

Key Milestones & Checkpoints

Recommended Prerequisites

Personal Study Notes

Video Tutorial

Official Guides & Documentation

Further Reading & Standards

Related EncodeEdge Tutorials

Keyboard Shortcuts

Reset All Progress?