Idea Details

Team and Contact Details

SNo Student Name School Degree Year Email
1Amna MasoodSEECSUndergraduateFourth14beeamasood@seecs.edu.pk
1Zainab TareenSEECSUndergraduateFourth14beeztareen@seecs.edu.pk
1Hamza ShafqaatSEECSUndergraduateFourth14beehshafqaat@seecs.edu.pk

Inter School Idea ? No
Do you need expertises from another area: No
If Yes please provide details of expertises you need:

Idea Details

Idea Name: ML-on-Chip
Slogan: Don't just develop, deploy!
Supervisor Name: Faisal Shafait
Supervisor Designation: Assistant Professor
Supervisor School: SEECS
Supervisor Department: DOC
Contact number: faisal.shafait@seecs.edu.pk
Email ID: 03330544462
Abstract:
    The project aims at efficient deployment of deep learning models in real-time embedded systems, accounting for the limited power and resource constraints associated with embedded systems
What is the unmet need in society that your idea will fulfill ?
    Currently there are 9 billion interconnected devices, which are expected to reach 24 billion devices by 2020.MI&S believes that the “machine learning” will account for a great deal of the innovation in IoT world. Providing local organizations with latest models for security will be our focus.
Who needs it ? How many would benefit ?
   Efficient implementation of CNNs in real-time security and surveillance systems will strengthen the security of organizations. It will prove to be a powerful hardware platform for all deep learning applications, especially safety-related, like face recognition, vehicle and area surveillance etc.
How will the solution works
    Deep Learning model will be efficiently implemented in C/C++ using Vivado HLS. Different optimization techniques will be applied in order to increase throughput and, at the same time, reduce latency and power consumption. These optimization techniques comprise of several strategies like minimizing memory accesses, reducing precision, pipelining and unrolling the structures carefully as not to exceed resource usage constraint,approximating computations,etc for attaining desired performance.
Who are your competitors ? How is your solution different
    Currently, there are a very limited number of products available for readily deploying Deep learning on hardware. This is a relatively new dimension in industry, born by the merging and recent surge in deep learning and Internet of Things. In Pakistan, this potential area has been poorly focusedupon
Status: new
Entry Date & Time: 2017-12-01 (0056)