Identifying vehicle accidents and high risk drivers using Machine Learning

The primary objective of the project is to approach the problem of understanding true causality of vehicle accidents and scientifically determining which vehicles and drivers are at highest risk of an accident from a machine learning perspective. Geotab has a number of identified collisions in X, Y and Z planes, and much more. The research would be aimed at using both Geotab’s data in addition to external data such as weather and topography to develop a predictive model that can identify those drivers at highest risk of an accident.

Demonstration-Based Initialization of Reinforcement Learning Algorithms for Efficient Robotic Control

Kindred’s Sort product is a robotic system that operates in e-commerce distribution centers to sort and handle apparel and general merchandise. The deployed system is controlled through a combination of artificial intelligence and human-in-the-loop teleoperation. The proposed project involves applying techniques from artificial intelligence (specifically machine learning and reinforcement learning) to improve the ratio of automatic control to human control.

Exploration of Methods and Models to Achieve Multi-Document Comprehension in the Legal Domain

The project attempts to tackle an important challenge in Artificial Intelligence (AI), to give a machine an ability to comprehend multiple documents like humans do. These can do the redundant or preliminary reading-based research performed in many domains. The project aims to create a system which can read, understand, and answer queries and/or summarize multiple legal documents in a single shot.

Cognitive Risk Sensing Using Deep Learning

CRISP is an international Deloitte development initiative aimed at helping some of our largest clients understand and managed corporate risk. CRiSP stands for “Cognitive Risk Sensing”, and it centers around using large sources of mostly unstructured data (i.e. 10% sample of all of Twitter, thousands of news aggregators, etc.) to understand and forecast risk for the clients. The goal for the student is to apply new methods in machine learning, data mining and natural language processing to extract user opinion of products from social media and customer feedback.

Voice Cloning Optimization

An artificial intelligence tool that is capable of generating natural-sounding speech can be embedded into many valuable services such as conversational agents for the disabled and conversational assistants. Such tool, when equipped with the capability of mimicking individuals’ vocal characteristics, will improve personalization of these services. In this project the intern will seek to develop a state-of-the-art voice cloning model.

Segmentation of 3D microscopy images

In-vivo imaging provides a unique opportunity to examine complex cellular activity in live tissue. Images produced by these experiments are difficult to analyze manually, typically applied to mono-layer cell culture assays (i.e. cells in a dish). Recent advances in deep learning enable the opportunity to analyze these in-vivo tissue images with greater efficiency and accuracy. This project will apply deep learning based segmentation and classification technology to a dataset provided by a collaborating pharmaceutical company.

An Artificial Agent for Light Switch

Smart home devices with artificial intelligence (machine learning and deep learning) will change our lifestyles in the near future. The objective of this project is to develop an artificial agent, which will power the smart light switches produced by ecobee. The artificial agent, a machine learning program, will use the data collected by the sensors in the smart light switches and help the users operate the light switches without the users’ manual control. The goal of this project is to develop an underlying smart program to learn the behaviors and of users with the light switches.

Flexible Data Reader on Distributed File Systems for Training Deep Learning Algorithms

With the fast-growing size of machine learning datasets, it has become increasingly important to store them in a reliable and distributed manner. Large scale distributed file systems such as GFS, HDFS and Amazon S3 have the capability to store large scale of data reliably. However, these distributed file systems have an intrinsic shortcoming: they provide good read/write access guarantees only for large size files, and therefore cannot efficiently handle frequent read/write operations for large number of small files.

Knowledge Intensive Processes Representation and Analysis: Process-Aware Work-Graphs and Predictive Approaches

Processes are important concepts in modern society since they control and standardize the interactions between businesses, consumers, governments and other organizations. However, the rise of knowledge-based industries such as financial services, healthcare, advanced manufacturing and software development have produced unstructured and knowledge-dependent processes. These Knowledge Intensive Processes (KIPs) KIPs range from partially structured to unstructured processes and require some control and standardization while guiding but not completely constraining knowledge workers’ actions.

UV Mapping Assistance through Deep Learning

The goal is to create a conversation loop between 3D designers and artificial intelligence programs. This will help the AI provide suggestions to the designer, while the designer provides the AI with feedback. This can help make it easier for designing complicated objects as well as complicated textures that belong to the surface of 3D objects. Through this interaction, the hope that AI can extend the utility of design software.