Machine Learning Projects for Final Year Students
Machine Learning Projects for Final Year Students
The final year project requires you to spend a certain amount of time producing a unique solution. You will research multiple model architecture, use various machine learning frameworks to normalize and augment the datasets, understand the math behind the process, and write a thesis based on your results.
16. Multi-Lingual ASR With Transformers
In the Multi-Lingual ASR model, you will fine-tune the Wave2Vec XLS-R model using Turkish audio and transcription to build an automatic speech recognition system.
First, you will understand the audio files and text dataset, then use a text tokenizer, extract features, and process the audio files. After that, you will create a trainer, WER function, load pretrained models, tune hyperparameters, and train and evaluate the model.
You can use the Hugging Face platform to store the model weights and publish web apps to transcript speech in real-time: Streaming Urdu Asr.
Image from huggingface.co
17. One Shot Face Stylization
In the One Shot Face Stylization project, you can either modify the model to improve the results or finetune JoJoGAN on a new dataset to create your stylization application.
It will use the original image to generate a new image using GAN inversion and fine-tuning a pre-trained StyleGAN. You will understand various generative adversarial network architects. After that, you will start collecting a paired dataset to create a style of your choice.
Then, with the help of a sample solution of the previous version of StyleGAN, you will experiment with the new architect to produce realistic art.
Image was created using JoJoGAN
18. H&M Personalized Fashion Recommendations
In the H&M Personalized Fashion Recommendations project, you will build product recommendations based on previous transactions, customer data, and product metadata.
The project will test your NLP, CV (Computer Vision), and deep learning skills. In the first few weeks, you will understand the data and how you can use various features to come up with a baseline.
Then, create a simple model that only takes the text and categorical features to predict recommendations. After that, move on to combining NLP and CV to improve your score on the leaderboard. You can also get better at understanding the problem by reviewing community discussions and code.
Image from H&M EDA FIRST LOOK
19. Reinforcement Learning Agent for Atari 2600
In the MuZero for Atari 2600 project, you will build, train, and validate the reinforcement learning agent using the MuZero algorithm for Atari 2600 games. Read the tutorial to understand more about the MuZero algorithm.
The goal is to build a new or modify existing architecture to improve the score on a global leaderboard. It will take more than three months to understand how the algorithm works in reinforcement learning.
This project is math-heavy and requires you to have Python expertise. You can find proposed solutions, but to achieve top rank in the world, you have to build your solution.
20. MLOps End-To-End Machine Learning
The MLOps End-To-End Machine Learning project is necessary for you to get hired by top companies. Nowadays, recruiters are looking for ML engineers who can create end-to-end systems using MLOps tools, data orchestration, and cloud computing.
In this project, you will build and deploy a location image classifier using TensorFlow, Streamlit, Docker, Kubernetes, cloudbuild, GitHub, and Google Cloud. The main goal is to automate building and deploying machine learning models into production using CI/CD. For guidance, read Machine Learning, Pipelines, Deployment, and MLOps tutorial.
Image from Senthil E
