Artificial Inteligence & Optimization The purpose of these practices was to evalute and compare huristic optimization method with neural network method(ANN)
- Task B- Gradient Descent: Numpy/Manual Calculations.
- Task C- Gradients using difference finite difference ratios.
- Task D- Auto-differentiation with TensorFlow2.
- Task E- Auto-differentiation with PyTorch.
This practice aimed to develpe a simple algorithms. PSO file was developed by Mohammed Al-Jaff.
In the first step, 10 particles were generated randmoly. Then those particles were modified and updated based on the two formula, and each time their loactions
were visualized by file_PSO. each time they approximately approached toward the optimal spot (minimum spot in curve)
- Task A - 1D Linear and non-linear regression - using simple FANNs.
- Task B: 2-variable regression - using FANNs to fit data drawn from a bell curve. (without validation and test set)
- More specifically, the is task to fit the parameters of a number of FANN architectures to a dataset D = {(xn, yn)} where each 2-dimensional input xn
is drawn from a uniform distribution on the square [−5, 5] × [−5, 5] and the response values (our yn) are determined as
y(x) = e^ (−xT x/4) = e^(−||x||^2/4)
Thus the task to train a FANN so that for a new input xn = [x1, x2], it will output a prediction of yˆn very close to the desired value.
- Task C: Signal Classification with FANNs.
In this task, you are given a dataset containing a fairly large amount ”signaldata”, each signal being labelled either as “normal” (numeric label 0) or
“abnormal” (numeric label 1). You assignment is to construct a FANN that is able to achieve a high classification accuracy when trying to classify a
signal as normal or abnormal. This mmakes this task a 2-class classification problem.
- Task 4: Image Classification of FashionMNIST using FANNs
- The dataset:well-known Fashion-MNIST1 dataset of Zalando’s article images consisting of 70000 images of clothing items in gray-scale: a training set
of 60000 examples and a test set of 10000 examples. Each example is a 28 × 28 grayscale image and associated with a label from one out of 10 classes.