[JdeRobot] GSoC in JdeRobot

[Rohan Saxena]

Hello mentors and developers,

I'm Rohan Saxena, an undergraduate sophomore studying Computer Science at
BITS Pilani, India (one of the top 3 engineering schools in the country and
the best private engineering college in India). I'm extremely interested in
machine learning and deep learning, and wish to contribute to JdeRobot by
implementing the deep learning project
<http://jderobot.org/Collaborate#Project_.238:_Deep_Learning_to_detect_objects_at_images_from_RGBD_sensors>
mentioned
on the ideas page.

Here I have been a member of the Embedded Systems and Robotics lab since my
freshman year itself. I work on robotics and artificial intelligence
(specifically computer vision and deep learning).

I am also a student at the Udacity self-driving car nanodegree, and as part
of the programme I have implemented various deep learning architectures
(especially CNNs) with established personnel in machine learning (PhDs,
scientists, professionals, etc). This has enabled me to develop a solid
understanding of the math behind ML algorithms like convolutions, for
instance.

For this project, I plan to employ the techniques mentioned in this paper
<https://arxiv.org/pdf/1507.06821.pdf> to use CNNs in RGBD object
recognition. I am fluent in implementing computer vision algorithms in C++
in TensorFlow.

I want to provide an architecture which can be directly deployed on a local
machine with pre-trained parameters (if the user wants), with options for
further training and optimisation and/or editing the architecture itself. A
user will be able to treat the architecture as a black box and load it to
train/test it as he/she desires. This will allow the user to load
pre-trained parameters trained on the standard dataset and train/test it on
a dataset of their choice.

Since the source code of the models will be available to the user, they can
also edit the existing model and tweak the hyperparameters, or the
architecture itself (add/edit layers, for instance), to try to increase
performance. They can then use this modified architecture as they wish.

Kindly let me know your views.

Sincerely,
Rohan Saxena
CS Sophomore
BITS Pilani, Pilani campus
Rajasthan, India
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://gsyc.urjc.es/pipermail/jde-developers/attachments/20170305/a05c1a81/attachment.html>