Mr Ashish Sundar
Masters in Engineering in Electrical and Electronic Engineering at Imperial
Thesis: Adversarial autoencoders for joint source channel transmission of videos
I mixed autoencoders and GANs to produce adversarial autoencoders to transmit videos over a noisy channel. The challenge would be to encode the videos in such a way that they take up less space but are also resilient to corruption via noise.
I then worked at National Physical Laboratory as a research scientist. There I worked on a few projects, some notable ones being:
- Using bespoke convolutional neural networks to classify image transformed ECGs
- Interpreting a bespoke remote sensing convolutional neural network designed to classify land use
- Researching self-reporting neural networks that provide a measure of their own uncertainty
- Using recurrent and convolutional neural networks to predict electric car battery charging and discharging curves given current current and voltage readings
I’m a first year PhD student working within environmental intelligence. I’m trying to get drones to work with each other using reinforcement learning so that they can map a previously unknown environment efficiently and quickly. I hope to use this in the Amazon rainforest to help the locals there begin an ecotourism project.
I am eager to learn more about reinforcement learning by applying it and am curious to see whether multiple agents operating drones operating in a partially observable environment can learn to communicate and cooperate with one another to accomplish given tasks.
I am also curious to see the potential of bleeding edge technology to improve society and provide a platform for those that need it. In this respect, I want this work to benefit communities and people at the lowest level, rather than doing so for large corporations. However, integrating this level of technology into small communities is challenging as it has the potential to disrupt their society, so I am keen to see how technology of this caliber can be integrated well into small communities.