Aishwarya Ganesan
Room 7361, Computer Sciences, W. Dayton Street, Madison
e-mail : ag at cs dot wisc dot edu
Home   Education   Work Experience   Courses   Publications   Projects  

Research Projects in Mobile Systems and Wearables

NVC: Hidden Communication Between Videos and Smart Glasses
We designed and implemented Near Vision Communication (NVC) that uses the visual link between a display device and the smart-glasses camera for transferring data. We built a system that embeds hidden information into video frames; while this information is imperceptible to human eyes, it can be extracted when the viewer watches the video through smart glasses.

Enabling Physical Analytics in Retail Stores Using Smart Glasses
We built a system that would enable the tracking of physical browsing by users in indoor spaces such as retail stores. Using a combination of first-person vision and inertial sensing using smart glasses, we track physical behaviors like walking, dwelling, gazing and reaching out. We also use the data gathered from smart-glasses to infer the product layout of retail stores.



Course Projects

A File System for Tiered Storage
We extended an existing file system (BPFS) to manage tiered storage consisting of non-volatile memory (NVM) and disks, using NVM as primary storage for hot data while using flash/disk as secondary storage for cold data.

ToyAFS
We built a simple distributed file system with crash consistency mechanisms to ensure that server or client or application crashes or adverse network conditions don't cause inconsistencies either to the server side file system or the client side cache.

Projects during Masters

Cost Based Query Optimization for Massively Parallel Query Processing Systems
We designed and built a cost-based query optimizer that provides an optimized plan for queries written in declarative languages built over systems like MapReduce by taking distributed execution into account. The query optimizer is based on the Cascades framework and was extended for Hive queries run on Hyracks (a data parallel platform to run data intensive jobs on a cluster)

Building Statistics for Hyracks/font>
Hyracks is a data parallel platform to run data intensive jobs on a cluster. We added the code of maintaining statistics into the Hyracks framework (an open source massively parallel system) by collecting relation level, partition level and attribute level statistics.

Answer It for Yahoo!
We developed an application that provides recommendations to users to answer unanswered questions on Yahoo! Answers by matching users’ interests with the content of the question. (This won Judges special mention at Yahoo HackU 2012)

Incremental View Maintenance in PostgreSQL for Aggregation
We modified the query processing logic in PostgreSQL to incorporate the feature of materialized views and provided support for materialized view definition and maintenance for aggregation based views.

Part of Speech Tagger using HMM
We implemented a POS Tagger and stemmer using Hidden Markov Model with bi-gram probabilities on British National Corpus. We applied heuristics and achieved an F-Score of 96.17%.