Benefit: This is an opportunity to create domains with out-of-range parameters (which you can’t execute in micromagnetic simulation). 

To extract parameters from and/or generate new images of magnetization patterns using micromagnetic simulation data, the images of simulated magnetization patterns will be input into Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs) for training.

Brain-computer interfaces represent a life-changing possibility for disabled patients to regain control of their lives, for example through highly functional prosthetics or movement devices. However, everyday movements, like walking, and the increased brain activity it causes pose significant difficulty for BCI decoders. 

This paper proposes a method to clean EEG data and create a decoder that is resistant to walking-induced artifacts so left and right-hand motor imagery can be classified. We first perform Artifact Subspace Reconstruction to remove major artifacts, then bandpass in the alpha and mu bands to find physiologically relevant frequencies, Common Average Reference Filter to improve spatial resolution, extract trials and samples, find the mean percent difference in power spectrum density between resting and motor imagery to best identify event-related desynchronization, perform fisher score feature selection to reduce dimensionality, and finally train and cross-validate our model offline and test online with evidence accumulation.

The paper is attached under publications. 

Some of the code for this work is available at https://github.com/zkhodzhaev/ML_brain_computer_interface

• Implemented the project by integrating TensorFlow and Keras, showcasing the ability to adapt and modernize legacy code with current deep learning frameworks.

• Constructed a neural network model using Keras with a hidden layer of 200 neurons, demonstrating skills in neural network architecture design.

• Employed Policy Gradient methods for intelligent decision-making, reflecting an understanding of advanced reinforcement learning techniques.

• Implemented custom preprocessing and a tailored loss function to optimize the agent's performance, displaying proficiency in data manipulation and algorithm customization.

• Demonstrated the agent's gameplay in real-time through OpenAI Gym's environment, showcasing the practical application of machine learning in a dynamic setting.

• Successfully trained, saved, and reloaded the model for demonstration purposes, illustrating the project's adaptability and reproducibility.

• GitHub code can be found here: https://github.com/zkhodzhaev/reinforcement_learning

• The study utilized multi-layer perceptrons (MLPs) for their versatility with diverse data types and relationships. MLPs with 50 and 100 nodes in hidden layers were chosen to balance complexity and efficiency, considering the feature sizes of datasets like the German Credit Dataset (27 features), Credit Defaulter Dataset (13 features), and Bail Dataset (18 features). This aimed to capture patterns effectively in datasets with 1,000 to 30,000 entries and prevent overfitting in smaller datasets. More details about datasets can be found here.

• MLPs were configured with various hidden layer sizes and activation functions, including 'identity,' 'logistic,' 'tanh,' and 'real,' ranging from simple to complex structures. Hyperparameters like learning rate and batch size added complexity and affected training dynamics and generalization capabilities. The study used default settings for these parameters to minimize variability.

• Dimensionality reduction techniques like PCA and t-SNE were implemented to manage high-dimensional data. PCA transformed features into principal components, preserving essential information, while t-SNE retained local data structures in reduced dimensions. PCA was iteratively applied to explore the impact of dimensionality reduction on model performance. t-SNE, limited to three components for efficiency, was used to assess its effect on MLP performance.

• The study showed variations in MLP performance across datasets. For instance, an MLP with a single 100-node layer achieved an AUROC of 76.02% and an F1-score of 80.28% on the German Credit Dataset. Fairness metrics like SP and EO varied, indicating potential predictive biases. Training MLPs on data with altered sensitive attributes slightly improved fairness metrics but reduced performance, emphasizing the need for techniques balancing fairness and predictive accuracy.

• Overall, the research highlighted the importance of careful model configuration and appropriate dimensionality reduction in ensuring fairness and stability in representation learning. 

• GitHub code can be found here: https://github.com/zkhodzhaev/representation_learning/