Aryan Malewar

• Data Scientist with 1+ years of experience executing data - driven solutions with adept knowledge of Data Science, Machine Learning (ML), Deep Learning (DL), Computer Vision (CV).
• Experience in fine tuning pre-trained models like Stable Diffusion from Hugging Face Model Hub.
• Experience with Data Manipulation packages like NumPy, Pandas, and Data Visualization packages like Matplotlib, Seaborn and Bokeh.
• Good knowledge of the Machine Learning Life Cycle including data preparation, model engineering, model evaluation, model deployment, monitoring and maintenance.
• Developed and experimented with various predictive and classification models, such as Decision Tree, Naive Bayes, Logistic Regression, SVM, Random Forest, Ada boost, Gradient boosting, Social Network Analysis, Cluster Analysis, and Neural Networks using Python and Scikit-Learn and assessed the performance using appropriate metrics and methods.
• Experience with state-of-the-art facial recognition methods and technologies like CNNs, R-CNNs, Faster R-CNN, and VGG-Face.
• Proficient in Python programming language as well as Object Oriented Programming in Python Hands-on experience with industry-standard IDEs like VS Code, Spyder, PyCharm.
• Good knowledge of cloud computing services like Amazon Web Services (AWS). Experience with version control tools like Git.
• Good communication and interpersonal skills, active team member and a smart worker.
• Experience in using project development and test management tools like JIRA, Confluence.

Neural Style Transfer (NST):

• Applied the neural style transfer technique to create artistic images by combining the content and style of different images using the pretrained convolutional neural network VGG-19 model.
• Extracted the features of the style images using a feature extractor model and defined a loss function to measure the difference between the features of the generated image and the original images.
• Implemented the model using the TensorFlow framework and optimized the generated images iteratively by stochastic gradient descent.
• Experimented with recurrent convolutional layers (ConvLSTM2D) and hyperparameters and observed their effects on the quality and diversity of the generated images.

Stable Diffusion XL Panaroma LoRA:

• Fine-tuned and applied the state-of-the-art Stable Diffusion XL panorama model from Hugging Face to generate realistic and seamless 360-degree panoramic images from text prompts using diffusion-based generative models.
• Used a custom dataset of 360-degree coherent panoramic images and their depth conditioning to fine-tune the diffusion model using the PyTorch framework and diffusers, autoencoders, and huggingface_hub libraries.
• Optimized the memory and speed of the training using gradient checkpointing and mixed precision and experimented with various text prompts and hyperparameters to improve the quality and diversity of the generated images.
• Compared the results with Stable diffusion model for 360-degree panoramic image generation and demonstrated the potential of the fine-tuned diffusion model.