• Probabilistic / Statistical Machine Learning

Probabilistic machine learning (such as Gaussian process regression/classification) focuses on building models that incorporate uncertainty. This approach allows for robust decision-making and inference by explicitly accounting for the inherent uncertainty in data and model prediction. On the other hand, statistical machine learning focuses on the theoretical aspects of learning algorithms including generalization analysis which evaluates how well a model performs on unseen data. 

• Collaborative Learning

Collaborative machine learning involves multiple parties jointly training a machine learning model while keeping their data decentralized and private. It aims to improve model performance by pooling knowledge from local data. This topic includes horizontal and vertical federated learning and knowledge distillation-based federated learning.

• Neural Processes

Neural Processes aims to solve regression problems in a stochastic manner when given a small amount of data with less than O(n^3) costs - which popular Gaussian processes take to play the same role. Like many of the neural models, it is a key factor to encode a given data so that the model can represent its given task. On the other hand, correlating the target inputs with each other should be considered as well from the stochastic estimator's point of view.

• Generative Models

A generative model is a type of machine learning model that learns to generate new data samples that resemble a given dataset by capturing its underlying distribution. These models, such as  GANs (Generative Adversarial Networks), VAEs (Variational Autoencoders), and diffusion models, can create realistic data examples like images. Our scope includes the generalization analysis of (conditional) GANs and decentralized supervised learning with missing data using generative models.