Birshreshtha Noor Mohammad Public College: Lecturer ICT: Written Exam: 2024প্রশ্ন ও সমাধান

Definition: Deep Learning is a subset of Machine Learning that uses artificial neural networks with many stacked layers ("deep" networks) to automatically learn hierarchical patterns directly from raw data, without manual feature engineering.

graph LR
    I["Input Layer<br/>(raw pixels/words)"] --> H1["Hidden Layer 1<br/>(edges/basic features)"]
    H1 --> H2["Hidden Layer 2<br/>(shapes/combinations)"]
    H2 --> H3["Hidden Layer 3<br/>(high-level concepts)"]
    H3 --> O["Output Layer<br/>(prediction/class)"]

Why "deep" matters: Each additional layer learns increasingly abstract features -- e.g. in image recognition, early layers detect edges, middle layers detect shapes, and later layers detect whole objects (faces, cars) -- something a shallow (1-2 layer) network cannot capture as effectively.

How it differs from classical ML: Classical ML often needs a human to hand-craft features (e.g. "edge count", "color histogram") before a simple model can classify data; deep learning learns the features itself from large labeled datasets, at the cost of needing much more data and compute (often GPUs).

Common applications:

• Digital assistants and voice-activated remotes (speech recognition)
• Automatic facial recognition and image tagging
• Fraud detection in banking transactions
• Machine translation and chatbots (NLP)

Exam takeaway: Deep Learning = Machine Learning + multi-layer neural networks + large data/compute; it powers most modern AI applications by learning features automatically rather than relying on hand-engineered ones.