Machine Learning and Its Tools

What is AI?

Artificial Intelligence (AI)
Definition: AI is a broad field of computer science that aims to create systems that can mimic human intelligence. It can include rule-based logic.

---

What is Machine Learning?

ML is a subset of AI that enables machines to learn from data without being explicitly programmed for every task.
Goal: Make predictions or decisions based on data by training algorithms.
It learns patterns from data, is always data-driven, and is more focused and technical.

---

How is it different from Data Mining?

Actually, data mining is when we extract some meaning from the data.
When we extract data, we then give it to a human, and the human is responsible for whatever informed decisions are made.

Machine learning is used as a small part — a tool — in this process.

But usually in machine learning, we are not making or extracting meaning for a person, but we are extracting meaning that the machine can understand. The machine extracts the meaning from the data and also tries to understand it in order to do some extra work on the same data.

---

Example: How a Machine Extracts Meaning from an Image

Let’s try to understand how a machine extracts meaning from a static image.

It’s just pixels, and pixels just have colors (red, blue, green). But the machine finds some sort of combination and then tells us what it actually is.

---

What Tools Are Used for Machine Learning?

Torch

It is an open-source tool which is built on Lua. It is a Python-like language. It is very powerful.
Torch is extremely scalable. You can run it on a cluster or multiple GPUs. It is easily scalable with just a change of a couple of lines of code.

If you write code for GPU and want to use the same code for CPU, you only need to change a few lines of code — not more than that — and it will be successful.

Modern GPUs today use the Torch library. We can write GPU-accelerated code using the Torch library.

---

Theano

Theano is basically an alternative model based on Python. It is symbolic computation.
Theano is a Python library (like PyTorch or TensorFlow) that was made for:

• Deep learning using symbolic computation

• First building a "math graph" of your model

• Then optimizing and running it efficiently on CPU and GPU

In Theano, when you define a model or calculation:

• You’re not doing it immediately

• You’re just describing the steps (like writing a math formula)

Then later, Theano:

1. Builds a graph of your operations (symbolic)

2. Optimizes it

3. Runs it efficiently on CPU or GPU

---

Note: First understand — Torch is dynamic, which means as we write the code, it runs the same way.

On the other hand, Theano is symbolic, meaning it builds computations first, then runs them.

But Theano is hard to debug and learn, and it is replaced by TensorFlow, JAX, and PyTorch.

---

Now, let’s discuss this further:

• Theano is replaced by JAX (it is symbolic graph-based but more optimized than Theano)

• TensorFlow is both symbolic and dynamic

  • TensorFlow 1.x is static graph

  • TensorFlow 2.x is dynamic by default

Dynamic execution makes coding and debugging easier.

Still, Python is slow for such tasks, so what does symbolic execution actually do?
It creates a backend C code, and then it runs that C code — which makes Python code faster. It does all the backend computation internally.

---

Keras

Keras is a high-level deep learning library in Python that lets you build and train neural networks easily.

You don’t need to write complicated code — just a few lines to define layers, and Keras handles the rest.

When you write code in Keras today, it’s really using TensorFlow under the hood.

---

Scikit-learn

Scikit-learn (or sklearn) is a Python library used for machine learning, especially traditional ML — like decision trees, SVMs, clustering, and regression. It is very well documented.

Scikit-learn is the go-to tool for classical machine learning — it’s fast, simple, and great for real-world datasets like CSV files, customer data, etc.

Where we need traditional ML (no GPU support), and for data types like CSV, tables, NumPy — Scikit-learn is very easy to use.
On the other hand, PyTorch and TensorFlow are for deep learning — where we need GPU support and work with tensor data types. These are more complex.

---

Mahout (Hadoop)

Apache Mahout is a machine learning library built to run on big data systems like Hadoop.

Mahout lets you run ML algorithms on huge datasets that can’t fit on a normal computer — using Hadoop's distributed power.

---

Example:
Imagine you have data from 10 million customers.
With Mahout + Hadoop:

• It splits the work across many machines

• All machines work together

• You get results fast, even for huge data

---

How does it run on many machines?
Suppose we have very big data — say, 5000 GB. If we want to do classification or apply some ML task on that data, it will take too much time.

What Mahout does:

• Breaks the data into chunks (e.g., 1000 GB each)

• Runs it on 5 different servers

• Each server processes its chunk using ML algorithms

• Results are collected and merged from all machines

We get the final output fast — even for big data.

---

Caffe

Caffe (Convolutional Architecture for Fast Feature Embedding) is a deep learning framework made by the Berkeley Vision and Learning Center (BVLC), focused mainly on image-related tasks like classification, detection, and segmentation.

It is used for:

• Image classification

• Object detection

• Running fast pre-trained CNNs

• Embedded systems (older edge AI projects)

Caffe is a deep learning framework made for fast image processing, especially CNNs. It’s fast and good for deployment, but less flexible — that’s why today most people use PyTorch or TensorFlow instead.

---

Final Note

A library gives you tools — you decide how to use them.
A framework gives you structure — it tells you how to build.

P.S.

If you spot any mistakes, feel free to point them out — we’re all here to learn together! 😊

Haris
FAST-NUCES
BS Computer Science | Class of 2027

🔗 Portfolio: zenvila.github.io

🔗 GitHub: github.com/Zenvila

🔗 LinkedIn: linkedin.com/in/haris-shahzad-7b8746291
🔬 Member: COLAB (Research Lab)