Constraints in Software-Based Machine Learning.

This short post will discuss the first principle constraints in software-based Machine Learning(ML) approaches. ML algorithms are broadly classified into supervised ML, unsupervised ML, and reinforcement ML. Let’s first talk about the first two categories, which mainly deal with “extracting meaningful science from a set of abstract data points”. There are two ways to increase the accuracy of the results; a) Increase the number of data points, or b) Build a better neural network, where the financial resources limit the former, and the latter is limited by the (wo)man power you have in your team. Given these two constraints, our job is “how to extract more science by using fewer data points than the existing framework”. Let’s apply limits on these two constraints. Financial constraints are about the money you can spend on this project, which can be all money available in the world (around $100 Trillion), and (wo)man power is limited by the number of people available in this world (around 7.7 Billion), in a hypothetical world where you can use all money and every living person is expert ML-engineer/mathematician.

Yes, it is almost impossible to make every living person (from infant to old age) literate in those skills but let’s take the most optimistic case scenarios by assuming the most positive limits on the given two constraints. This is the universe from which you have to pick up the combination of the values of these constraints for any given ML problem in the world. If you have limited financial constraints that can buy you only 10 data points at maximum, then your job is to find a better ML/mathematician who can help you extract a little more science from the existing 10 data points. If you cannot find a better ML/mathematician, then find ways to get more money to generate 11th, 12th … data points until you get the desired results. As far as the third ML type (reinforcement ML) is concerned, it is all about second constrained, i.e., the number of people in the world, and less about financial constraints. However, some hybrid techniques can also use a set of data points, but that will be a secondary part of that framework. For this approach, we are limited by the brain power of the most expert person in the world in the given reinforcement ML domain, and/because we can not generate a new human being for the purpose of building better ML code.

I hope you enjoyed reading this, and thank you for stopping by. Have a good day/night.     

Materials as a proof of concept (MaaP)

During my today’s paper reading session I had an epiphany about the usability of the 2D materials or in fact any material for that matter. I came across a paper [1]. That says, there was a person 50-ish years ago who theoretically devised a quantum phenomenon (Exciton Condensation) and in this paper, we are giving experimental proof of that phenomenon in a 2D material (MoS2). From reading this a thought came to my mind since they could have used any other 2D materials for this purpose so materials do not have to serve the only purpose of application in a device they can serve as a platform to give the proof of a concept about a theoretical phenomenon that has not been given yet. Or we can go one step further and say since there are infinite configurations that exist in which atoms can be arranged to form discovered/undiscovered materials there must be infinite theoretical concepts that shall exist which will be experimentally confirmed by those unrealized materials.

[1]

https://assets.researchsquare.com/files/rs-940264/v1_covered.pdf?c=1633111041

Quantum computing.

Computers works on the principle of 1s and 0s. So there is only 2 position an event can take. Wich will be replaced quantum bits(qbits) in quantum computers. Silicon Valley, Wall Street, and Academia are equally excited about this. Two quantum computer development stage companies have recently went public i.e. IonQ and Rigetti. IonQ is scientifically backed by academics from the University of Maryland and Duke University[1].

Recently one of the major journals “Nature Nanotechnology” decided to dedicate their December issue cover to the quantum computing aspect of nanotechnology[2]. Although Nanotechnology exist for almost 50 yrs now while recent push by academics to quantum computing seems to be fueled by the commercial aspect of the sub-field application. As today Gene Munster, a managing director at Loup Ventures, validated the field’s potential saying “All this EV adoption fueled by Tesla is very good but quantum computing is gonna make it look like appetizer” on the show Closing Bell with Will Frost on CNBC channel. Gene’s further comments on the field can be seen on their YouTube channel[3].

[1] https://ionq.com/company

[2] https://www.nature.com/nnano/volumes/16/issues/12

[3] https://www.youtube.com/watch?v=Lr-ymZXeQ1I&ab_channel=LoupTV

Minorities of the whole world in American academic institutions.

I am writing this post aimed at an audience who are not from India and wants to interact with a person from India and have little idea about do’s and don’ts about social dynamics in that country as I did in an exactly opposite sense when I moved to States from India. Although I am primarily considering that audience from the USA, it can be generalized for any other country. In the USA, the color of one’s skin is a significant factor in deciding who will be categorized as majority and who will go into the minority category. Please keep in mind; I said it is a “significant factor” & not “THE factor”. It is a somewhat binary system, i.e., white Vs. non-white. A similar social hierarchy exists in India in addition to one’s skin color. And that is the cast system. It’s a kind of spectrum instead of a binary system. On the highest end of the spectrum, those people exist who were employed by doing the “noble” works in the castle of the ancient kings, i.e., accounting, religious affairs, chef, white-collar jobs, etc., recognized as “Forward Classes by the Govt. of India” currently. And on the lower end of the spectrum, people are categorized into another category who were involved in doing the “dirty” works for the king, i.e., king’s mistress, sewage/toilet maintenance, etc., recognized as “Backward Classes by the Govt. of India”. And just to give an idea of social dynamics, even today, in some parts of India, the Backward Class people are called “untouchables” as slang because a person from the Forward Class does not feel “comfortable” shaking a hand with a person from the backward class. And in between these two categories, an almost semi-spectrum exists. But let’s, for the moment, just focus on the endpoints to make the story easily digestible. There is nearly a significant correlation between the Castes and the skin color, while it is not exclusive. Still, it holds true for the majority of the time, i.e., the lighter the skin color a person has, the higher the chances that (s)he belongs to the one of the Forward Class and darker the skin color then higher chances (s)he belongs to the Backward Class.  

Next time, if you are in a higher position in academics and want to give an opportunity to a person from India, please consider the person’s background in addition to the qualification. Because a person from India will be viewed as non-white in the USA and will be given the opportunities reserved for the minorities here, while that person is really a majority from where (s)he is coming and lived like it their whole life. And (s)he will never speak out about it and will quietly accept the opportunities that (s)he does not deserve because those are reserved for people who lived their life as a minority (no matter in what part of the world) and (s)he very clearly knows what (s)he is doing. Since the people in the USA do not know about the cast systems of India so this will keep on happening and hence this post. And non-Indian-Americans will fall for it because skin color is the primary factor here, and Indian-white-skin=American-non-white-skin. Next time you hire a Ph.D. student, post-doc, research scientist, etc., from India in your lab; please be curious to know that candidate’s background from this perspective and if you want to score some extra points under your “underrepresented categories are encouraged to apply” initiatives.

And following is a little note for the people who belong to the Backward Class and moving to the States and have to deal with the people from the Forward Class in high positions, which is usually the case since most of the successful Indian-American people belong to the Forward Class. Since you have to deal with them, there are a few points I want to tell that I hoped somebody would have told me when I moved to the States: 1) Comes in terms with your background, never fool yourself into believing you will not be marginalized in the Free World based upon your caste. 2) If you have dark skin color, then you will be marginalized by the non-Indian-American and Indian American both, and if you happen to be one of the few who are from the Backward Category and still have light skin color, then you will be marginalized by your own people i.e. Indian Americans. Because unfortunately, they will always make sure you never forget your background when you stand in front of them because **they know it**. Do not let these 2 points leave your mind until the day you die. I am a Backward Class Ph.D. candidate at my work surrounded by Forward Class people from India, and I am writing this not from any point of hate, none so ever. For the American audience, that experience will be like a black person surrounded by white colleagues at work. I believe every white colleague will encourage that black person to share their experiences and support him/her wholeheartedly, no matter whether the experience was negative or positive. And there will be lots of things that people will do without even realizing how hurtful it can be for that black person. For example, I am a first-generation Ph.D. student in my extended family, and my family says it with pride that somebody from **them** is not only going to college but earning a Ph.D. degree. And when I told this with an identical sense of pride to one of my colleagues, the reply was, “why your parents will let you go to college if they didn’t go themselves” and walked away while smiling. At that time, I didn’t take it to my heart because I thought I was one of them because caste-based institutionalized marginalism does not exist in this country, and it was light-hearted humor. But now, if I look back in hindsight, it impacted my mental health, and I hope writing it here will help me. I am sharing this one experience to be cathartic to somebody like me reading this if this ever happened to you, and you shall be prepared for such interactions if this didn’t happen to you yet. Again, do not reply with any kind of hatred when such interactions occur. Just hear it and move along with your job and use it as constructive motivation, and share it with your “Tribe” down the line, if anything.    

Please be considerate while reading this post that English is not my 1st language, and sometimes I use the wrong words to convey my message. And I need your help to fill the gaps while reading this if something does not come across right. And if somebody is reading from my workplace, please be informed it is not hatred but a helping hand for another person like me who will be entering the academia since we know the human brain tends to pose better mental health if more relatable experience exists out there in the world. And this is one of my relatable experiences for my “Tribe”.   

And again, the main aim of this post is not for any kind of hatred towards anybody but educational posts for the American people (especially academics) to make them aware of this point of view. A recent visit to India by Twitter CEO Jack Dorsey was primarily focused on this issue[1].

I wonder what “caste systems” exist in other countries in addition to skin color and money. If you are from a country other than India and USA, please comment down if you know a similar social hierarchy in your country. Thank you for your time and efforts in reading this post; with love.

[1] https://www.theguardian.com/technology/2018/nov/20/twitter-ceo-jack-dorsey-criticised-for-upsetting-hindu-nationals-india-visit

How to find your dream job

Do one simple thought experiment check on your job. Find yourself a work, that you will still do if 7.7 Billion people of the world (the world’s total population) is against you. You shall not be afraid of doing it even then. And on the other hand side, if you feel afraid of doing something even if only one person out of 7.7 Billion people in the world is against you then stay away from that work.

Our brain on “freezing”

According to psychological science, there is no bigger fear than the fear of the unknown as of today’s research body. A day/month/year/century from now somebody can prove that there are fears which are bigger than fear of the unknown.

We always tend to the things that we know instead of the things that we don’t know. Now we can easily apply limits on the things that scare us or make our brain freeze in fight/flight/freeze response situations. Our brain freezes when we start assuming an infinite number of unknown things can happen to us. And on the other hand, if you want to un-freeze your brain then you need to believe that the number of things that can happen to me is not infinite. Even if that thing is a bad thing. You can think it like this, the fear created by severe bad things but finite in number are much lesser than the fear created by mildly bad but infinite things. That’s why people say whenever you are feeling afraid, just think what is the worst outcome and suddenly you start feeling better because the moment you think about the worst outcome, exactly at that moment you tell yourself that there are only finite outcomes instead of infinite, even if they are bad.

Something bigger than Artificial Intelligence.

Recently my mom visited me from India and I thought to visit a local historical building with her to see the beautiful architecture of the building so that she won’t feel homesick as she goes out. And it is a very big building with a tall roof. While sitting inside and looking at a tall standing roof, I had suddenly start thinking about every human that is living in the world. And next thought came is, can I put every human being in the volume of the building. Then I thought that it may not be possible so I did one thought experiment. Just to consider that human part of the body that makes them most unique i.e. brain. So, if I measure the volume of the average human brain and then multiple that with 7.7 billion units, the answer may be not exactly matched with the volume of the space inside that building but it will be closer i.e. may 10 church buildings will have more volume than the number of brains inside the world.

Then I thought to myself, that is it. This amount of volume is running everything in the world right now. Following that, days were passed but that thought experiment was somewhere sitting in the back of my head. And then I realize what will happen if we can put all of the human’s brains next to each other and connect them somehow, will they outperform a single brain? Let’s go one step ahead, let’s say we found out a way by which we can connect these brains so that if we connect two brains then they will outperform 1 brain by the factor of 2. And the strongest brain we can make is joining all of the living human being’ brains in the world right now. Please keep in mind it was(is) just a thought experiment. I have a background in material science and I know theoretically it is possible to make any material in the world, given you just have to arrange atoms in the proper positions in the space. So, this led to my next question, instead of using brains developed inside a human body, why not we can build a synthetic brain that will be equivalent to the human body in terms of its IQ or whatever parameter you want to consider. And then we can create 7.7 Billion such brains and line them up in a way that the 2 brains lined up will be stronger than the one brain by exactly the factor of 2. Now just imagine a person/nation/organization that can build these synthetic brains and line them up will have the advantage of the brainpower equivalent to the people in the world.

Now as a next step, two variables can be considered; first- how to lineup the 2 brains in such a way the resultant power will be more than by the factor of 2 and second- if we can arrange the atoms in such a way that we can create a synthetic brain then why not go ahead and find out another arrangement of the atoms so that just one synthetic brain will be equivalent to the 7.7 billion human brains. There will be no stop to it. If one country is able to achieve the 7.7 billion to 1 ratio then some other will work to make a let’s say 20 billion to 1 ratio synthetic brain and so on, there will be no end to it. And a building housing this synthetic brain will be protected with 10 times more security than the Pentagon.     

Almost a year later after my 1st thought experiment in the church, I came across a podcast hosted by Lex Fridman. In episode #250, a person named Peter Wang appears in it[1]. And he talked about an idea that is almost similar to what I was entertaining in my mind. Please be advised, this idea was not completely minted by me. I already know that artificial brain neurons are a work under progress as I come across several scientific papers about Memtransistors under my habit of reading out loud new papers for 1 hour every weekday as they are delivered by my google scholar alerts.  But I never realized, it has this much potential so that the computer-based AI will look nothing in front of it. It is not the question of yes or no, it is just the question of when. And once that happened then there will be no way to put back the genie in the bottle and in fact, there is no way to stop the genie from coming out from the bottle in the first place.

[1]

https://youtu.be/X0-SXS6zdEQ?t=3262

Update #1 (Jan 17, 2021): Just came across the Lex Fridman podcast episode[2], where a person named Jeffrey Shainline from NIST (Boulder, CO, USA) is a guest and talked about a similar idea followed by his recent paper[3]. He is calling it “Optoelectronic Intelligence”.

[2]

https://youtu.be/EwueqdgIvq4

[3] J. M. Shainline, “Optoelectronic intelligence,” Appl. Phys. Lett., vol. 118, no. 16, p. 160501, Apr. 2021.

Grain boundaries in overlayer on a substrate

Let’s assume there are only two ways to get grain boundaries in the universe of grain boundaries in layers materials. First, the overlayer is feeling strain due to the existence of the substrate under it. Second, due to multiple nucleation sites result in multiple domains which results merging of those domains and hence the grain boundaries.

In below paper, the authors are working on the second reason by neglecting the first on the basis of interaction between substrate and overlayer is too weak to produce any strain in overlayer due to lattice mismatch or any other reason.

Now let’s apply the limits on these scenarios and see what we get. First case, let’s assume the overlayer was grown by only a single nucleation site and hence single grain and hence no grain boundary. And starting from 0 interaction between overlayer and sublayer, we are increasing the interaction gradually to infiniti. A question can be asked in this hypothetical scenario of increasing the interaction gradually, do we reach a point (before hitting infinity) where the single-crystal-overlayer will start experiencing a distortion in its crystal in a way to have more than one crystal. If yes then what is that threshold. We know from our real-life experience that this threshold do exist and it is way less than infinity because whenever we grow overlayer in a lab over substrate it HAS multiple domains and hence grain boundaries.

On the second grain boundary reason, we can setup a hypothetical scenarios like following. For a given interaction between substrate and overlayer, do at least one combination of all growth parameters exist (among all their permutations and combinations) so that we will have more than one nucleation site but the grains originating from all nucleation sites will have SAME orientation and no grain boundaries!* A let’s take it one step further, how this combination varies as a function of the interaction between substrate and overlayer.

I got the above thoughts from this paper https://aip.scitation.org/doi/abs/10.1063/5.0057417

A few days after writing this blog post I came across this paper[1]. It almost demonstrates the same idea i.e. they observed orientation dispersion of MoS2 resulting into wafer scale MoS2.

Thank you for stopping-by, have a good day/night ahead.

[1]

https://www.sciencedirect.com/science/article/abs/pii/S0169433221018626

Training, Validation, Test Data/Set in Machine Learning Algos

In case if you are wondering what is difference between above three or especially validation and test data/set, as I was. Please follow the link below for a simple explanation.

https://stats.stackexchange.com/a/130289/338040

P.S: I have no association with the account who posted that comment but found it useful for me. Hopefully it will be for you too. Thanks for stopping by.