Artificial Neural Networks (317531)

T.E. AI & Data Science, SPPU 2019 pattern. Covers biological foundations through modern deep architectures — perceptrons, backpropagation, CNNs, RNNs, and transfer learning.

Instructor: Tushar R. Mahore

Term: Even Semester (AY 2025–26)

Location: ICEM, Pune — Department of AI & Data Science

Time: As per SPPU timetable

Course Overview

This course provides a systematic study of Artificial Neural Networks — from biological inspiration to modern deep learning architectures. By the end of the course, students will be able to:

  • Understand the theoretical foundations of neural computation
  • Design and train feedforward, convolutional, and recurrent networks
  • Apply neural networks to classification, regression, and sequence problems
  • Use frameworks (TensorFlow / Keras) to implement and evaluate models
  • Interpret model behaviour using activation maps and visualization tools

Pattern & University

Savitribai Phule Pune University (SPPU) T.E. Artificial Intelligence and Data Science — 2019 Pattern Subject Code: 317531 · Subject Type: IN (Internal)

Lab Component

Practical sessions cover implementation of key architectures using Python (TensorFlow / Keras / PyTorch). Students are expected to submit a term work journal with all practicals completed.

Evaluation

As per SPPU pattern:

  • Internal Assessment (Theory): 30 marks
  • End Semester Exam (Theory): 70 marks
  • Term Work (Practical): 25 marks
  • Oral / Practical Exam: 25 marks

Contact

Email: mahoretushar@gmail.com · Office hours by appointment.

Schedule

Week Date Topic Materials
1 Unit I Introduction to Neural Networks

Biological neuron model, McCulloch-Pitts neuron, activation functions, perceptron learning rule, limitations of single-layer perceptrons.
2 Unit II Multilayer Perceptron & Backpropagation

Feedforward networks, gradient descent, backpropagation algorithm, vanishing gradient problem, weight initialization strategies.
3 Unit III Radial Basis Function Networks

RBF architecture, learning methods (supervised, unsupervised), comparison with MLP, applications.
4 Unit IV Self-Organizing Maps (SOM)

Unsupervised competitive learning, Kohonen SOM, neighbourhood functions, applications in clustering and dimensionality reduction.
5 Unit V Recurrent Neural Networks

Elman and Jordan networks, BPTT, vanishing gradient, LSTM and GRU architectures, sequence modelling applications.
6 Unit VI Deep Learning & Convolutional Neural Networks

CNN architecture (convolution, pooling, fully connected), training deep networks, batch normalization, dropout, transfer learning, pretrained models (VGG, ResNet).