This research will promote the development of machine learning strategies that can be applied toward general cognitive traits, measured in different ways, across different populations.
In this effort, PIs Alina Zare and Paul Gader are developing machine learning approaches to translate sensor data across sensor platforms for a unified analysis framework.
This proposal uses operators from quantum theory to define uncertainty in machine learning decisions as well as in data streams. The core innovation is to use a modal decomposition of the probability density function that quantifies better the distribution tails.
This proposal enhances the conventional machine learning architectures with novel memories inspired by the cortical columns that form the mammalian cortex. The goal is to model cortical columns as automata that implement inference as a search process. This module will be integrated with our current cognitive architecture for autonomous vision.
Reinforcement learning (RL) is very slow and does not scale up well to high dimensional spaces. This proposal employs causality to go beyond correlation learning to reduce the amount of data necessary to train RL. Pointwise interactions with the environment will be substituted by trajectories to speed up training, while the value of information optimizes the exploration exploitation dilemma.
This proposal seeks to improve the architecture of the conventional state model of control theory by extracting events from the flow of time and storing them in specialized independent memory banks. The theory of the value of information will be employed to select the events and reduce the memory capacity.
As emerging and future computing systems are heading toward “extreme heterogeneity”, the focus of this project is to study the application of heterogeneous computing (CPUs + GPUs + FPGAs + emerging memories) to accelerate machine learning and other data analytics applications.
The goal of this project is to address these gaps via two orthogonal approaches: design of more efficient algorithms for learning, and development of scalable exploration techniques that can lead to efficient learning.
Prof Meyn is developing new approaches to reinforcement learning (RL) and distributed control with wide-ranging applications. This project concerns the development of this theory, and its application to reliable control and resource allocation for demand dispatch (an alternative to demand response for smart grid applications).
This CAREER project will advance the state of the art in supervised machine learning to allow for incomplete, uncertain and unspecific label information. Supervised machine learning algorithms produce desired outputs for given input data by learning from example training data. The methods generally rely on completely and accurately labeled training data to drive the learning algorithm. However, many applications are plagued with labels that are incomplete, uncertain, and unspecific (lack precision). Current techniques do not adequately handle such data.