Machine-Learning-Zero2Hero
A series of notebooks that introduce Machine Learning concepts with hands-on practice and its mathematics in brief.
Posts by Collection
portfolio
Interpretability Analysis of Bilingual Brain-Informed Fine-Tuning in BERT
This project investigates how bilingual brain-informed fine-tuning changes BERT’s internal computations. We analyse BERT models fine-tuned to predict fMRI responses from bilingual participants reading naturalistic English and Chinese stories, comparing them with their pretrained base counterparts across syntactic structure, representational similarity, language-specific token routing, factual retrieval, and the causal roles of individual layers and attention heads. Our central question is: what changes inside a language model after bilingual brain-informed supervision?
Deep-Learning-Fundamentals-with-TensorFlow
A series of notebooks that introduce Fundamental Deep Learning concepts with hands-on practice using TensorFlow.
Brain Tumor Segmentation using 3D U-Net
A work on segmenting brain tumors on BraTS Dataset using Pre-trained and Fully-trained 3D U-Nets with commonly used loss functions to assess the effect of loss function on generalization error.
Covid-Data-Analysis
A simple data visualization and analysis of Daily Covid data provided by John Hopkins University.
Biofunctions
This repository contains Python implementations of bioinformatics algorithms, including several that are featured in the Rosalind problem set.
DeepLabV3Plus-TF-Backbones
This project extends support for DeepLabV3+ implementation on TensorFlow with multiple backbones, including: ResNet50/101/V2, DenseNet121/169, MobileNet/V2, and VGG16/19.
CNNs-Introduction
A series of notebooks that provides an in-depth introduction to Convolutional Neural Networks and the most common CNN architectures with hands-on practices.
StoMNeuro
Final project for the Stochastic Modeling in Neuroscience course at TU Berlin, involving a PyTorch implementation and evaluation of neuLVM for mesoscopic modeling of hidden spiking neurons.
PCP-Hitman
After AlphaZero’s impressive success in the arguably hardest combinatorial two player games, like Go and Chess, we want to find out whether it can help with yet another great challenge…connect four! :) This repository contains the code to train an alphazero like model in connect four, as well as the option to play against different agents (alphazero, random, minimax, mcts) or to let them play against each other.
publications
Design of Neuron-specific Lentiviral Vectors and Investigation of Gene Therapy Applications in Spinal Muscular Atrophy Disease
Published in 2023
This poster is a summary of my thesis work for the Molecular Biology and Genetics department that I presented during my defense.