The purpose of this repository is to make prototypes as case study in the context of proof of concept(PoC) and research and development(R&D) that I have written in my website. The main research topics are Auto-Encoders in relation to the representation learning, the statistical machine learning for energy-based models, adversarial generation networks(GANs), Deep Reinforcement Learning such as Deep Q-Networks, semi-supervised learning, and neural network language model for natural language processing.

GAN encoders in PyTorch that could match PGGAN, StyleGAN v1/v2, and BigGAN. Code also integrates the implementation of these GANs.

This repository contains my solutions for the Coursera course TensorFlow: Advanced Techniques Specialization. Expand your knowledge of the Functional API and build exotic non-sequential model types. Learn how to optimize training in different environments with multiple processors and chip types and get introduced to advanced computer vision scenarios such as object detection, image segmentation, and interpreting convolutions. Explore generative deep learning including the ways AIs can create new content from Style Transfer to Auto Encoding, VAEs, and GANs.

Recent deep generative models are able to provide photo-realistic images as well as visual or textual content embeddings useful to address various tasks of computer vision and natural language processing. Their usefulness is nevertheless often limited by the lack of control over the generative process or the poor understanding of the learned representation. To overcome these major issues, very recent work has shown the interest of studying the semantics of the latent space of generative models. In this paper, we propose to advance on the interpretability of the latent space of generative models by introducing a new method to find meaningful directions in the latent space of any generative model along which we can move to control precisely specific properties of the generated image like the position or scale of the object in the image. Our method does not require human annotations and is particularly well suited for the search of directions encoding simple transformations of the generated image, such as translation, zoom or color variations. We demonstrate the effectiveness of our method qualitatively and quantitatively, both for GANs and variational auto-encoders.

Used a Generative Adversarial Network(GAN) to simulate a optical communication system; Applying auto-encoding deep-learning method to build an Intelligent model of Communication system in order to improve its information rate and symbol coding method; laying a good foundation for the development of intelligent fiber communication systems;

Used the Functional API to built custom layers and non-sequential model types in TensorFlow, performed object detection, image segmentation, and interpretation of convolutions. Used generative deep learning including Auto Encoding, VAEs, and GANs to create new content.

Built a deep learning model to generate 3D sign poses of ASL using OpenPose, Progressive Transformers (encoder/decoder architecture) and GANs. Automation of sign language production benefits deaf community.

Variational Auto Encoders (VAEs), Generative Adversarial Networks (GANs) and Generative Normalizing Flows (NFs) and are the most famous and powerful deep generative models.

A deep learning system to combat audio deepfakes by embedding a speaker's voiceprint (from an Encoder) as an inaudible watermark (using a GAN).

Used the Functional API to built custom layers and non-sequential model types in TensorFlow, performed object detection, Image segmentation, and Interpretation of convolutions. Used generative Deep Learning(Unsupervised Learning) including Auto Encoding, VAEs, and GANs to create new content.

This project employs advanced deep learning techniques, particularly Generative Adversarial Networks (GANs) for realistic image generation and transformation. Using the pSp (pixel2style2pixel) model, based on StyleGAN, it performs latent space encoding to enable fine-grained control over attributes like age, smile, and hairstyle in facial images.

Explaining and predicting behavior from neural activity has been a longstanding goal in neuroscience. It is known that visual information is encoded within the hierarchical structure of the visual cortex, and plays an essential role in visual processing. However, the decoding of stimulus images from neural activity is still a challenging topic. Here we ask whether neural activity in the visual cortex of mice can be used to decode stimulus images, and whether specific visual cortex subregions recreate the images better than others. We hypothesize that neurons in the primary visual cortex (VISp) would best recreate these images. To investigate this, we employed a decoding approach outlined in previous literature. We obtained image visual features from a pre-trained deep residual neural network (ResNet), and created a linear mapping to corresponding neural activity (spike counts). This was then used to reconstruct the stimulus images through a generative adversarial network (GAN)-type layer. We observed that our model successfully decoded stimulus images from neural activity within a 70% accuracy. In addition, we found that VISp neurons achieve greater decoding quality relative to other subregions (80%). We conclude that our model can be used to accurately reconstruct stimulus images from neuronal spike counts, and that neuronal activity in the VISp encoded the majority of the information. Our findings may inspire a simple yet effective architecture for novel brain-computer-interface applications. Since our dataset contained a limited number of images and neuronal responses from one subject, generalization may be limited. We also have not examined whether combinations of subregions can recreate images better than single

Deep ANN, CNN, Auto-encoder, RNN, GAN, Transfer Learning in PyTorch

Deep learning project for colorizing Synthetic Aperture Radar (SAR) images using PyTorch and GANs. Transforms grayscale SAR images into realistic RGB representations using encoder-decoder architecture.

Functional API, Advanced computer vision scenarios such as object detection, image segmentation, and interpreting convolutions. Generative deep learning, Style Transfer to Auto Encoding, VAEs, and GANs.

No description available

Deep learning models in the field of communication: Encoder, Auto-encoder, GAN

The main purpose behind this lab is to get familiar with Pytorch library, to build deep neural network architecture for Auto-encoder, Variational Auto-encoder and GANs (Generative AI).

The main purpose behind this lab is to get familiar with Pytorch library, to build deep neural network architecture for Auto-encoder, Variational Auto-encoder and GANs (Generative AI).

The main purpose behind this lab is to get familiar with Pytorch library, to build deep neural network architecture for Auto-encoder, Variational Auto-encoder and GANs (Generative AI).

Deep learning pipeline for restoring damaged celebrity portraits using U-Net Autoencoder, GAN, and Hybrid Context Encoder. Includes masking pipeline, custom losses, and evaluation metrics.

Using Deep Convolutional GAN's discriminator and Encoder of AE as a feature extractor to classify DomainNet images with a random forrest (the "real" domain))

This project implements a Context Encoder for semantic image inpainting using a deep encoder-decoder architecture. It reconstructs missing regions in images by learning from visual context, using a combination of masked L2 reconstruction loss and adversarial loss (GAN) for realism.

A deep learning project that generates natural language captions for images using a pre-trained EfficientNet CNN and Transformer-based encoder-decoder architecture. Trained on Flickr8k with data augmentation and GAN-enhanced samples.

For this side hustle project, I built and integrated 3 deep learning models, designing 2 main ones: 1) Customized CNN Model with ~83% accuracy 🤌, and 2) GAN Model with ReLU Activation & One-Hot Encoding, achieving ~89% accuracy 🤌.

A deep learning–based image steganography system that hides secret messages without embedding them into existing images. Inspired by the PLOS ONE paper “Image Steganography Without Embedding by Carrier Secret,” it uses GANs to generate realistic carrier images encoding hidden information.

a deep learning system to detect audio deepfakes by embedding speaker voiceprints as inaudible watermarks. Designed a CNN-based speaker encoder and a 1D-CNN GAN to embed a 64-bit voiceprint hash, implementing a twostep verification workflow using Python and PyTorch.