First Blog of Team No-9

Project Topic: 2D Oscillators for ONNs to solve CSPs

Team Members: Sai Gaurav, Vishwas Bharti

Mentor: Shashwat Shukla

Objective: Use 2D n- Oscillator Neural Network as an extension to 1D n-ONN, with the required Cost function to solve Constraint Satisfactory Problems like the Travelling Salesman (TSP).

Implementation:

As a part of 1st week, we had read the paper “n-Oscillator Neural Network based Efficient Cost Function for n-city Traveling Salesman Problem” by Shruti et al. A synopsis of this paper is as follows:

CSPs like the Travelling Salesman are NP hard problems, difficult to be solved by digital computers/algos. Alternatively, Neural Networks(NNs) with appropriate cost function can be made to solve such problems with parallel processing. Many other papers also suggested this idea, and designed appropriate NNs with the required cost function to be minimised. One interesting idea was to map it to oscillator units instead of usual neural activators and encode the problem in the phases, as implemented in one such paper with oscillators. The Cost function has been designed to map the optimal solution/route on to a unit circle with each city mapped to each root of unity, keeping in mind the invalid routes, taken care of by appropriate distance and syntactic constraints. The cost function is minimzed using Gradient Descent. To avoid getting stuck in a local minima, simulated annealing is used wherein stochastic noise that decayed with time has been added.

The complexity has further been reduced to the order of ‘n’ as implemented in this paper. Most of the constraints were similar i.e. to encode the problem onto the roots of unity onto a unit circle, with similar syntactic constraints, with a slight modification on the distance one wherein the weight was decided based on a gaussian function with decreasing variance, rather than absolute difference in unity roots as implemented in previous paper. A comparison has been made between both the implementations on the factors like computation, time of convergence, search space, hardware requirements, etc depticting the latter to be better in these aspects.

The code based on this implementation was also shared with us, which further aided in understanding of the paper. We did run the code and here are some of the results:

random coordinates of 10 cities

Output mapped onto unit circle with same symbols used: order depicting the optimal route

Further Plan: Dynamics of 2D Oscillators shall be thought of, after discussing with the Mentor and its implementation. The previous code shall also be rewritten in Python for future use in our project.

Work Report of Team No-6 for the week(01-03-2019 to 10-03-2019)

Project Title- Compact and energy efficient oscillators using Antiferroelectric phenomena (Direction-1), Vivek Saraswat

Last week we went through the papers provided by our mentor. One of them is Ferroelectricity(FE) and Antiferroelectricity(AFE) of Doped Thin HfO 2 -Based Films. It deals about the effects upon the FE and AFE of HfO 2 upon various conditions and also tells about the applications. There are further 2 more papers that needs to be read. Till now we are reading the above mentioned paper and hope to complete majority of papers by the next week.

Work Report of Team No-2 for the week(25-02-2019 to 03-03-2019)

Project Title-Thermal Modelling of PMO devices with Thermal Insulation Mentor-Sandip Lashkare

This week we have gone through the study material given by our project mentor. Firstly a research paper on effects of thermal insulation on NbOx switches as a preliminary to understand the effect of the same on PMO devices. Just the current expression changes due different mechanisms in both of them. To understand the same we went through a blogpost shared with us. Then we started with a paper by MA Lampert to understand the SCLC mechanism though we have not gone through it completely. We plan to finish it by the end of next week. Thats all for this week.

EE724 Project Updates

• The purpose of this blog is to have a continuous interaction with students, TA’s and Instructor with respect to project updates.
• Anyone from the project team writes the blog with team number and project title.
• Students shall post the project updates weekly, mandatorily. (Even if nothing is done in a particular week, they will post “no updates this week”, but they have to post)