What to Learn Next — AI & Deep Learning

Why This Course Focused on Classical Machine Learning

Everything you've built in this course — cleaning data with pandas, visualizing it with Matplotlib/Seaborn, and modeling it with scikit-learn's linear regression, decision trees, and random forests — falls under classical machine learning. It's not a lesser starting point; it's the toolkit that handles the large majority of real-world tabular data problems: spreadsheets, database tables, CSVs with rows and named columns.

Deep learning — models built from neural networks with many layers — is a different, more specialized toolkit. It didn't get skipped because it's unimportant; it got skipped because it solves a different kind of problem than the one this course spent 32 lessons preparing you for, and reaching for it too early is a common beginner mistake.

What Neural Networks Add — and When They Actually Help

A neural network is a model made of layers of connected "neurons," each layer learning to represent the data a little differently than the layer before it. Stack enough of these layers (this is the "deep" in deep learning) and the network can learn to recognize patterns far too complex to hand-engineer as features — patterns in raw pixels, raw audio waveforms, or raw sequences of words.

That power comes at a real cost: neural networks typically need far more data, far more compute, and far more tuning than a RandomForestClassifier does — and on small or medium tabular datasets, they frequently don't even outperform it.

A Taste of the Deep Learning Toolkit

If you do move toward deep learning next, here's the vocabulary and tooling you'll run into almost immediately.

🧰 The two dominant frameworks

Almost all deep learning code today is written in one of two Python libraries: TensorFlow (usually through its high-level Keras API) and PyTorch. Both are free, open-source, and capable of the same range of work — the choice between them is largely a matter of preference and what a given tutorial or team already uses.

# Keras (TensorFlow) — sequential, declarative style
from tensorflow import keras

model = keras.Sequential([
    keras.layers.Dense(64, activation='relu'),
    keras.layers.Dense(1, activation='sigmoid')
])
model.compile(optimizer='adam', loss='binary_crossentropy')

# PyTorch — explicit, class-based style
import torch.nn as nn

class Net(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc1 = nn.Linear(10, 64)
        self.fc2 = nn.Linear(64, 1)

🖼️ CNNs for images

A Convolutional Neural Network (CNN) is the standard architecture for image tasks. Instead of connecting every pixel to every neuron, it slides small filters across the image to detect simple shapes first (edges, corners), then combines them in later layers into more complex patterns (textures, then whole objects). This is the same core idea behind image classifiers, object detectors, and most computer vision work.

🤖 What a transformer actually is

A transformer is a neural network architecture built around "attention" — a mechanism that lets the model weigh how relevant every other word in a sentence is to the word it's currently processing, rather than reading strictly left to right. It's the architecture behind most modern language models, powering tasks like translation, summarization, and conversational AI assistants. You don't need to implement one from scratch to use this technology — most beginners start by using pretrained transformer models through a library rather than training one from zero.

Other Directions Beyond Deep Learning

Deep learning isn't the only path forward from here — plenty of people build entire careers on classical ML skills alone, applied more deeply in a specific direction.

How to Choose What's Next

There's no single correct next step, and no need to rush into deep learning just because it sounds more advanced. A few honest, practical guidelines:

Lesson Summary