Abstract

Motivated by the need to improve medication adherence of patients through life analytics, we consider the problem of recognizing human activities using accelerometer data from wearable devices. In this talk, I will first outline the challenges that we face when trying to recognize changes in human activities with real-world data. I will then focus on a baby step towards the larger goal: to detect changes in the activities performed by the patient. Existing change detection algorithms use either unsupervised methods that are dependent on a pre-defined threshold or supervised methods that learn from the data. The latter mostly focuses on simple ambulatory activities or uses data captured in scripted settings, making it difficult translate into real-world scenarios. I will then present our proposed change detection approach that primarily uses shape features combined with hierarchical clustering to tackle the complexity that arises from real-world settings.

Abstract

In this talk, I will outline a series of work on improving the viewer's experience in the context of DASH Video Streaming. I will first present a simple queueing model that we use to model a DASH player, relating the video buffer, playback rate, network throughput, and video bitrate. Then, I will show how this model is used as a basis for different algorithms for DASH, to improve the viewing experience in different contexts, including multi-server streaming, tiled streaming, and live streaming.

All faculties and students are welcome to join.