- Task: Building models for image classification, from scratch
- KNN, SVMs, Decision Trees, Random forests
- Experiments on CIFAR-10, CIFAR-100 dataset
- Accuracies and runtime for both tarning and testing
- w.r.t. a varying number of train data points
- Hyperparameter search with CV
- Accuracies and runtime for both tarning and testing
- Task: Building models for regression, from scratch
- Linear and nonlinear GP regression (vector-valued model)
- i.i.d. Gaussian noise model
- Isotropic Gaussian kernel (for nonlinear)
- Experiments on SARCOS dataset
- Accuracies and runtime
- Accuracy surface w.r.t. varying hyperparameters
- Computationally efficient GP regression using the subset of regressors (SOR) approximation
- Comparisons with other regression algorithms
- Other kernels
- Accuracies and runtime
- Task: Writing a survey paper on Neural Processes
- Academic publications discussed in the paper:
- Neural Processes(NPs)
- Conditional Neural Processes(CNPs)
- Attentive Neural Processes(ANPs)
- Sequential Neural Processes(SNPs)
- The Functional Neural Process(FNPs)
- Task: Performing neural network regression algorithms on 3D Hand Pose Estimation,
- Experiments on the 3D Hand Pose Estimation data
- Compare models
- RF, GP, MLP, CNN, and more
- Accuracies and run-times
- Details of hyperparameter selection
- Ablation studies
- Compare models