[mlpack] GSoc 2019 - mlpack : evolutionary algortihm project proposal idea discussion

[Naman Gupta]

Hello Marcus,

Thank you for your positive response. I am working on the proposal now, and
I will focus on the recommendations made by you. I will send you the
proposal for review in 2-3 days. I am hoping to work with you under your
guidance this summer as a part of GSoC 2019. Thank you for your time and
consideration.





On Thu, Mar 21, 2019 at 3:46 AM Marcus Edel <marcus.edel at fu-berlin.de>
wrote:

> Hello Naman,
>
> thanks for the clarification. I like the idea and I think this could fit
> well
> into the mlpack codebase. One point that you should focus on in your
> proposal is
> how this could be implemented so that it integrates well into mlpack; what
> methods can be reused what has to be added.
>
> Let me know if I should clarify anything.
>
> Thanks,
> Marcus
>
> On 19. Mar 2019, at 20:58, Naman Gupta <namangupta0227 at gmail.com> wrote:
>
> Hi Marcus,
>
> First of all, thank you for your quick response. In my previous email, I
> mentioned all the algorithms I have implemented in my research work. They
> all perform well as feature selection methods (filter and wrapper both)
> when compared to the traditional Grid search and Random Search algorithms.
> As per your suggestion, I would like to point out that out of them, *Grey
> wolf algorithm*, *Crow Search* and *Whale Optimization algorithm* perform
> significantly well on a number of datasets. I have tested their performance
> on a number of datasets as follows:
> 1. Parkinson's Dataset
>      a) Audio dataset
>      b) Image dataset
>      c) Text dataset
> 2. Thyroid dataset
> 3. Medical images
>      a) CT scan images
>      b) Microscopic images
>      c) MRI scans
> 4. Protein structure dataset
> 5. White Blood Cells images dataset
>
> *All these datasets have high dimensionality. Some of them have both small
> instance set + high dimensionality which is a great challenge in machine
> learning and has not been addressed effectively.
>
> According to a comprehensive survey by my lab and the aforementioned
> implementations, these 3 algorithms are most suitable for feature selection
> methods. In the last few years, bio-inspired optimization algorithms are
> recognized in machine learning to address the optimal solutions for complex
> problems in science and engineering. One of such problem: "High
> dimensionality problem" which in turn requires high computational time and
> affects the classification accuracy due to noisy features, redundant etc. Feature
> selection reduces the dimensionality of the data by eliminating features
> which are noisy, redundant, and irrelevant for a classification problem. It
> is most often a challenge for the researchers due to its computational
> complexity.
> So, I would like to model and implement these 3 algorithms as feature
> selection methods (filter based and wrapper-based both) in mlpack this
> year. The users will be able to modify the algorithm and will be able to
> define their own fitness functions and many more features.
>
> Please let me know what you think.
>
> Thank you for your time and consideration.
>
> Naman Gupta
>
>
>
> On Tue, Mar 19, 2019 at 2:59 AM Marcus Edel <marcus.edel at fu-berlin.de>
> wrote:
>
>> Hello Naman,
>>
>> welcome and thanks for getting in touch. I like the overall idea, but I'm
>> not
>> entirely sure which method you propose to implement; it sounds like you
>> like to
>> work on: Crow search algorithm, Grey wolf optimizer, Cuttlefish
>> algorithm, Whale
>> Optimization algorithm, Ant lion optimizer, personally I would focus on
>> one or
>> two methods that have a record to perform well on certain tasks. Let me
>> know
>> what you think.
>>
>> Thanks,
>> Marcus
>>
>> On 18. Mar 2019, at 16:30, Naman Gupta <namangupta0227 at gmail.com> wrote:
>>
>> Hello Everyone.
>>
>> I am Naman Gupta, a computer science undergraduate student at MAIT,
>> GGSIPU, Delhi, India. I have been working on bio-inspired evolutionary
>> algorithms for the past 2 years and I have developed and implemented
>> various optimized versions of different bio-inspired algorithms in various
>> fields including Ad hoc networks, Medical Image Processing, and NLP. Some
>> of my work has been published in SCI-indexed journals (Q1 ranking).
>>
>> I have been working on bio-inspired algorithms namely, Crow search
>> algorithm, Grey wolf optimizer, Cuttlefish algorithm, Whale Optimization
>> algorithm, Ant lion optimizer, etc. and their usability in various domains.
>> These algorithms are inspired by the social behavior of animals in nature
>> and provide far more superior results when compared to the state of the
>> algorithms (Evolutionary and Genetic algorithms). Bio-inspired algorithms
>> are gaining popularity day by day because of their capability of finding
>> solutions to NP-hard problems and are being applied to a myriad of
>> optimization engineering problems like Thermal design, Structural
>> optimization, Satellite layout design etc. I have the statistics of over
>> ten years representing the growing number of applications of these
>> algorithms. I have developed and modeled these algorithms as feature
>> selection algorithms (filter based and wrapper based) which finds the most
>> optimal feature subset from a large feature dataset. It resolves the “curse
>> of dimensionality” problem more efficiently and with less computational
>> time and higher classification accuracy. I have already implemented the
>> aforementioned algorithms in python during my research work.
>>
>> I am very much interested in contributing to mlpack in GSoC'19. Now, I
>> want to implement these algorithms in mlpack as feature selection mathods.
>> These algorithms are population-based, Meta-heuristic optimization
>> techniques and are simple, flexible, and avoids local optima. They search
>> the global search space in less computation time as compared to the
>> traditional approaches Grid search and Random Search. They will enhance the
>> classification accuracy and will reduce the computational time.
>>
>> It would be a great help if the mentors could provide me some insight
>> into this proposal idea. Can I propose this idea? Can you please suggest me
>> something to make it better. I will add more details, more functionality,
>> and features in the final proposal, this is just an abstract. I look
>> forward to hearing from you.
>>
>>
>>
>> _______________________________________________
>> mlpack mailing list
>> mlpack at lists.mlpack.org
>> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20190322/db990d81/attachment-0001.html>