Current and Upcoming Projects#
Term 1 Projects#
Interesting Title |
Boring Title |
Objective |
Skills |
Status |
|---|---|---|---|---|
Predicting house prices |
Improving a model’s performance without changing the model |
Improve the performance of a model without changing the model itself by cleaning and pre-processing the dataset, extracting meaningful features, performing hyperparameter tuning and implementing cross-validation. |
Data Cleanup, Data Pre-processing, Feature engineering, Feature impact analysis, Hyperparameter tuning, Cross-validation techniques |
|
Detecting Fraudulent Transactions |
Model Evaluation and Anomaly Detection |
Develop an anomaly detection system to identify fraudulent transactions using unsupervised techniques; whilst leaning how to effectively evaluate the model using metrics like accuracy, F1-score, and AUC. |
Anomaly detection, working with imbalanced datasets, precision-recall analysis, model evaluation metrics (accuracy, F1-score, AUC). |
ETA: Term 1 |
TBD |
Unsupervised Learning and Clustering |
Apply clustering techniques to segment customers based on their purchasing behavior or other characteristics. |
Unsupervised learning, clustering (K-Means, DBSCAN, hierarchical), cluster evaluation using silhouette score, determining optimal clusters with elbow method. |
ETA: Term 1 |
Detecting Cancer from CT Scans |
Data Augmentation and Synthetic Data Generation |
Develop a system that generates synthetic data or applies data augmentation techniques (e.g., image flips, rotation, cropping) to increase the size and diversity of a training dataset. |
Data augmentation techniques, synthetic data generation, enhancing model generalization, working with limited data. |
ETA: Term 1 |
Mitigating Bias |
Ethical AI and Bias Mitigation |
Analyze a machine learning model (e.g., for loan approval or hiring) for bias against certain groups and implement techniques like fairness-aware learning or adversarial debiasing to mitigate it. |
Bias detection and mitigation, fairness-aware learning, applying fairness metrics (demographic parity, equal opportunity), ethical AI considerations. |
ETA: Term 1 |
Solving Optimization problems |
Optimization and Metaheuristics |
Use a genetic algorithm or another metaheuristic optimization technique (e.g., simulated annealing) to solve an optimization problem like the Traveling Salesman or Knapsack Problem. |
Genetic algorithms, metaheuristic optimization, simulated annealing, designing fitness functions, solving complex optimization problems. |
ETA: Term 1 |
Term 2 Projects#
Project |
Real-world Application |
Objective |
Skills |
Status |
|---|---|---|---|---|
Building a Harry Potter Q&A Bot |
Building a pipeline to use pre-trained LLMs |
Using pre-trained LLMs and NLP methods to build a bot to answer questions based on input text (e.g. Harry Potter books) |
Natural language processing (NLP), working with large language models (LLMs), building conversational agents, fine-tuning pre-trained models, prompt engineering, text data preprocessing |
ETA: Term 2 |
TBD |
Explainable AI |
Train a complex model like a neural network or ensemble method and use explainability techniques (LIME or SHAP) to interpret its predictions |
Model interpretability, feature importance, explaining black-box models, ethical AI concerns |
ETA: Term 2 |
TBD |
Image Classification with Transfer Learning |
Use a pre-trained Convolutional Neural Network (CNN) (e.g., ResNet, VGG) on a new image classification task by fine-tuning the last few layers |
Transfer learning, convolutional layers, image augmentation |
ETA: Term 2 |
TBD |
Generative Models |
Create a Variational Autoencoder to generate new images by learning the underlying distribution of a dataset like MNIST or CIFAR-10 |
Generative models, VAEs, latent space interpretation, sampling from learned distributions |
ETA: Term 2 |
Training a robot cat to walk |
Sim-to-Real Transfer Learning |
Train a robot control model in a simulated environment and then adapt it to real-world conditions using transfer learning. |
Transfer learning, reinforcement learning, simulation-based training, domain adaptation |
ETA: Term 2 |
Medical Diagnosis |
Bayesian Machine Learning |
mplement a Bayesian neural network to estimate uncertainty in predictions, especially for tasks like medical diagnosis where confidence is crucial |
Bayesian inference, uncertainty quantification, Bayesian networks, handling out-of-distribution data |
ETA: Term 2 |