COM S 402 Week 8 Presentation By Malik

COM S 402 Week 8 Presentation By Malik Bulur, Tao Li, Timothy Schommer, Carter Wunsch, Junran Zhang

Problem being addressed ● Deep Learning/Machine Learning models require hyperparameter tuning to be useful ● Tuning hyperparameters nowadays is done by manual trial and error ● The underlying theory behind a model’s hyperparameters can be too complex

Solution: Genetic Algorithms ● An algorithm that shares characteristics with nature ● Used to solve optimization problems ● Algorithm Steps: 1. Create initial population 2. Fitness Function 3. Selection 4. Crossover 5. Mutation 6. Repeat steps 2 through 5 for certain amount of generations Binary Encoding: Value Encoding:

Design and Development Issues and Resolutions ● Learning new concepts and implementing our own unique solutions ● Refactoring code ● Genetic algorithm decisions ● Deep learning decisions

Development Practices & Tools ● ● Iterations in the form of solution testing and refinement, rather than progressive goals Tools: ○ ○ Python language Tensorflow Keras Numpy

Current Progress Status Reza Morsali

Current Progress Status Step 1: Learning Genetic Algorithm Step 2: Write Genetic Algorithm for One variable Step 3: Rewrite Genetic algorithm for multi variables Step 4: Rewrite Genetic Algorithm for Multi-Object Functions Step 5: Learning Sample Convolutional Neural Networks Step 6: Write CNN for classification problem

Current Progress Status Step 1: Learning Genetic Algorithm Step 2: Write Genetic Algorithm for One variable Step 3: Rewrite Genetic algorithm for multi variables Step 4: Rewrite Genetic Algorithm for Multi-Object Functions Step 5: Learning Sample Convolutional Neural Networks Step 6: Write CNN for classification problem

Current Progress Status Step 1: Learning Genetic Algorithm Step 2: Write Genetic Algorithm for One variable Step 3: Rewrite Genetic algorithm for multi variables Step 4: Rewrite Genetic Algorithm for Multi-Object Functions Step 5: Learning Sample Convolutional Neural Networks Step 6: Write CNN for classification problem

Current Progress Status Step 1: Learning Genetic Algorithm Step 2: Write Genetic Algorithm for One variable Step 3: Rewrite Genetic algorithm for multi variables Step 4: Rewrite Genetic Algorithm for Multi-Object Functions Step 5: Learning Sample Convolutional Neural Networks Step 6: Write CNN for classification problem Pareto frontier

Current Progress Status Step 1: Learning Genetic Algorithm Step 2: Write Genetic Algorithm for One variable Step 3: Rewrite Genetic algorithm for multi variables Step 4: Rewrite Genetic Algorithm for Multi-Object Functions Step 5: Learning Sample Convolutional Neural Networks Step 6: Write CNN for classification problem

Current Progress Status Step 1: Learning Genetic Algorithm Step 2: Write Genetic Algorithm for One variable Step 3: Rewrite Genetic algorithm for multi variables Step 4: Rewrite Genetic Algorithm for Multi-Object Functions Step 5: Learning Sample Convolutional Neural Networks Step 6: Write CNN for classification problem

Remaining Work Timeline Step 7 : Deep Learning: improving CNN for a data set ● ● Adding more layers Number of layers is at least 10

Remaining Work Timeline Step 8 : Integration: Genetic algorithm + Deep learning ● ● The goals: define objective function, choose the variables, code development. Variables from GA contribute to the response of the CNN

Remaining Work Timeline Step 9 : Integration: Adding more variables ● ● hyper parameters ○ convolutional layer: filter size, number of filters ○ max pooling: filter size. ○ Dense layer: number of nodes. total number of pair layers ○ ○ convolution + max pooling

Remaining Work Timeline Step 10 : Testing WITH Vermeer data set Step 11: Adding more complexity / Possible publication ● Try to enhance the technique and compare the performance with earlier works.

Demo ● ● Sharing Screen Focus on the GA problem and CNN