Indian IT Industry

Javee · Post by **Javee** » 10 Jun 2017 08:48

Soft bank picked up Boston Dynamics from Google. Not sure of the price, but not sure what's up with Alphabet,have they ever sold something before, that too in a growth area like AI?

KJo · Post by **KJo** » 10 Jun 2017 09:13

Marten wrote: Only IT scion who has proven any kind of mettle is Shiv's (Nadar, not our Dr) daughter. And even that is probably because of the running start? The big 4 will make a larger comeback in India by gobbling up one of the majors or the other.

She was my batchmate here in the US. But then she is lucky that because of big name and opportunity and mentorship unavailable to most others.

Marten · Post by **Marten** » 10 Jun 2017 09:49

Javee wrote:Soft bank picked up Boston Dynamics from Google. Not sure of the price, but not sure what's up with Alphabet,have they ever sold something before, that too in a growth area like AI?

Perhaps they decided to get out of defense?
Have seen big dog in the first year of development, strutting around their car park! Would have been an amazing investment for an Indian firm. Robotics moving from warfare to civilian applications will be a massive sector in the next decade. US MIC has been slower than China in that arena.
PS: Intuit was there right across the road. One of the prod managers moved to Alphabet.

Singha · Post by **Singha** » 10 Jun 2017 09:58

Infy founders wanting to sell their 12% stake is front page ET today. Denied by founders and corporate.

Could be potential buyers and proxies leaking a steady stream of bad news and rumours to drive down stock price ahead of talks

Can infy and wipro merge together ?

On netz and hpe merger...folks were saying layoffs would go from bad to "nuclear"

Javee · Post by **Javee** » 10 Jun 2017 10:40

Marten, absolutely TCS has a sky high pile of cash and has no vision. To their defense, don't know whether they are equipped to run a business like that.

But what does netz and hp have in common? Layoffs will be more from the support I assume??

negi · Post by **negi** » 10 Jun 2017 13:19

Singha wrote: Can infy and wipro merge together ?

That would be such a blood bath , remember when HP acquired EDS ? In the end there were no winners but only those who facilitated the M&A.

On netz and hpe merger...folks were saying layoffs would go from bad to "nuclear"

Even a foot soldier from trenches like me can say with confidence that this will be a failed marriage . Actually as a general rule of thumb behemoths never combine to deliver same volume of business as they would be doing when operating separately . Issue is 2 years will be lost in getting BUs aligned , redundancies identified and removed, attrition and churn to stabilize and most importantly GTM and Sales to get aligned to new entity to again start generating leads and converting these numbers.

Netz should stay away from large companies it has more to lose versus HPE , companies like HPE and their shareholders are hoping for likes of Netz to bail them out of their misery . If there is a space for viable server+network appliance stack play in the market no one is stopping them from entering a big JV hey but that will not please the financial activists ,M&A vultures and the shareholders

CalvinH · Post by **CalvinH** » 10 Jun 2017 21:34

HCL should be able to merge with Infy or Wipro. They are very strong in IMS space whereas Infy/wipros strength is in the ADM space. Lately IMS is under huge pressure and HCL has a bleak long term future. May be with the combined strength and some acquisitions the new company can move to cloud hosted DC space. HCL also has background in large DC management toot that Infy/Wipro lack.

Large M&A for creating shareholder value are rare anyways.

I think daughter is opening schools/Colleges and Hospital and not actively involved in IT anymore. For India Medical and Education are sunrise sectors with better long term returns.Grapevine was that HCL was planning to move large centers from tier-1 cities to tier-3 cities with their own hospitals, schools and colleges that can serve the employees. Give it from one hand and take from other.

KJo · Post by **KJo** » 10 Jun 2017 21:53

negi wrote:
Singha wrote: Can infy and wipro merge together ?
That would be such a blood bath , remember when HP acquired EDS ? In the end there were no winners but only those who facilitated the M&A.

On netz and hpe merger...folks were saying layoffs would go from bad to "nuclear"
Even a foot soldier from trenches like me can say with confidence that this will be a failed marriage . Actually as a general rule of thumb behemoths never combine to deliver same volume of business as they would be doing when operating separately . Issue is 2 years will be lost in getting BUs aligned , redundancies identified and removed, attrition and churn to stabilize and most importantly GTM and Sales to get aligned to new entity to again start generating leads and converting these numbers.

Netz should stay away from large companies it has more to lose versus HPE , companies like HPE and their shareholders are hoping for likes of Netz to bail them out of their misery . If there is a space for viable server+network appliance stack play in the market no one is stopping them from entering a big JV hey but that will not please the financial activists ,M&A vultures and the shareholders

You are all mistaken if you really believe the tripe that M&A is to "increase value" or "exploit synergies". The only thing it really does is enrich the the top execs who are responsibile. They make money and add to their resume. My previous company got bought twice in a year. It was engineered by a desi top exec who bought it. I happened to meet him in the gym a month ago and introduced myself. He is now CEO at another firm. Buying us added to his resume.

Very Chankian. Our CEO also left and became CEO of another organization. Employees left holding the bad and all top execs gone citing some lame excuse or other. CPO said "It is not that I am leaving this company, it is just that I am going to another place

". Luckily I got out, but some are still there.

Vikas · Post by **Vikas** » 11 Jun 2017 00:24

For a vast majority of S/W foot soldiers, all that would matter is continuity in Job and some salary hike.
In my KB, target is to cut staff by 20% to 30% in next 4 qtrs with hardly any fresh hiring.

Javee · Post by **Javee** » 11 Jun 2017 08:21

Any merger with Indian companies will result in a bloodbath for employees, not that the chairman's or CEO's care. I don't think infy will merge, Wipro has an outside chance but then in the last couple years they have lost customers in drove. Wipro used to be strong in certain industry verticals, like telecom which hasnt done well across the board. So, it will be a tough sell for Wipro. There was a rumor couple years ago that HCL will be bought over by Accenture, but then they hired top execs from Accenture to trim the salaries and assorted fat for the foot soldiers. HCL truly has gone in to tier 2 cities, operations have commenced in Madurai, Lucknow etc.

asgkhan · Post by **asgkhan** » 11 Jun 2017 10:17

Costs in the salt mines I work are controlled by not backfilling roles, no more comp offs, working on weekends is expected, travel banned and all training done on goto meetings and video conferencing.

Singha · Post by **Singha** » 11 Jun 2017 11:15

tit for tat reply seen in a layoff forum

<company name> should get out of India and start over in Pakistan. Pakstanis are more civilized and innovative than Indians.

...

Looks like you still have weekend hangover. But I don't blame you because your masters in Al-Q HQ need more western assets in your country to blow up.

SRoy · Post by **SRoy** » 11 Jun 2017 13:09

Javee wrote:Any merger with Indian companies will result in a bloodbath for employees, not that the chairman's or CEO's care.

Coupled that with broken corporate hiring process, social cost will be very high.

Suresh S · Post by **Suresh S** » 11 Jun 2017 20:26

Reading about the story of mathematician Anand Kumar in Patna and his super 30 is inspiring to say the least. I have no shame to admit that I for last few years have been making Khayali Pulav as to how one day I could do something exactly like this. But buffoons like me only think and heros like Anand Kumar actually have done it . Real hero of India, hats off to him.

CalvinH · Post by **CalvinH** » 12 Jun 2017 01:34

The labor arbitrage driven model has another effect. Tier-3/4 IT services companies are mushrooming in India and they are eating business from tier-1/2 Some of them are now doing business with IT in Fortune 50 companies. I know few companies who are doing $20-$100 Million business. They are like the glorified body shops but have been able to scale it to a decent level with defined service offerings and good offshore strength to back up these offerings. Have sales staff in US too. Non MBAs/Non engineers.

Indian rising to senior management in IT in US companies is making it easier for these Tier-3/4 companies as well. As pressure on reducing cost increases internally Indian IT leadership is much more amendable to switching providers to lower cost ones as the first reaction. Buys them more time.

These companies are providing jobs to lot of non engineering graduates back in desh. These companies are stand to gain initially.

KJo · Post by **KJo** » 12 Jun 2017 23:48

I was asleep on Saturday morning and at 7:30am I thought I heard my cell phone ring. It stopped so I went back to sleep and in a minute it rang again. It was a call from India! No name associated with it, and I heard it could be my parents wanting to talk to me urgently, so I picked up. It was a recruiter fellow saying he was calling from Flipkart, BLR. He had seen my resume somewhere (naukri?) and maybe that's how he had my cell number.

So I say I was asleep and it's 7:30 here and he laughs. So I ask him to send me an email instead. He says he will and he has my resume (which version???) and all he wanted to have was my expected compensation and "when I could move to India".

How nice, I thought. He wasn't even asking me if I was interested in the position or not. He could have emailed me first and waited for a reply but instead he called me over Whatsapp. And he wanted to know when (not if) I was moving to India.

His email reached me on Sunday and when Sunday night he sends me a Whatsapp message saying that he sent me the email. WTF I thought... Whatsapp is a personal thing, it isn't email. I don't want to be whatsapping people I didn't know. Then he called me again this morning when I was talking to my mother over the phone.

HR practices seem a bit different in India than here in the US.

Talking about HR practices, I interviewed over phone with a well known company in Seattle 3 months ago. The interviewer (HR screener) didn't seem to be interested in seeing if I was a match. He seemed bored and disinterested as if the original HR lady had forced him to do it. Then a week passed and I did not hear back. I emailed the first lady and no reply. Then a week later I email again and no reply.

So I was pissed, I wrote a stinker of a review on Glassdoor.com about my experience. The VP of HR saw my review and asked the reviewer to write to her and gave her email address. I did after a few weeks and sent her an explanation of what happened, with screenshots from jobvite of all communication. And that it still said "In Process" when 2-3 months had passed. She replied apologizing. Today I get a

email from the recruiter lady apologizing saying this wasn't the way they do things, and things got mixed up between her and the other guy.

Looks like VP memsaab went jihaadi on the recruiter. I usually don't like to get people in trouble because everyone's lives are so stressful anyway, but if you can be so careless and high and mighty that you cannot reply to 2 emails asking for closure, and you cannot change status of the interview to "Rejected" then you need an ass-kicking from your boss.

Vikas · Post by **Vikas** » 13 Jun 2017 15:58

KJo, You are lucky. I had experiences where the recruiter herself started interviewing me and advising me that I should not mention Money as the reason of Job Change. Another moron was trying to negotiate salary with me despite that they are outsourced folks.
Most of Indian Recruiters are people who could not find Job anywhere else and are forced to go thru the motions or are housewives.

Few weeks back I got a LinkedIn message from Micky Talent mgmt team if I would be interested in joining a new group of theirs and asked for my phone number and then silence...Never heard back from them.

KJo · Post by **KJo** » 16 Jun 2017 23:38

I think Harpreet Singh is overreacting.
Indian IT's business model is to pick whatever is "hot" in the market, train the vast armies of graduates in that technology and send them to the US and market them with a "buy 1 get 10 free" model. This has worked very well for years. I don't see why it won't work now.

In 1999 it was Java/C++ and Y2K. In 2017 it will be Artificial Intelligence and Machine Learning and Analytics.
Teach Machine Learning and Data Science to 21 year olds in a 2 month crash course, and exploit demand.

From Linkedin
Artificial Intelligence and the Death of Indian IT Sector

Over the last 25 years, India has emerged as an important destination for information technology—the low-cost counterpart for IT services that China is to low-cost manufacturing. But the low-end IT services niche that India occupies is highly vulnerable to the next phase of technological disruption in automation and artificial intelligence as novel capabilities such as self-repairing code reduce the need for the large-scale deployment of cheap IT professionals. Voice-enabled everything and increasing customization will make near-sourcing both practical and desirable which in turn lessens if not eliminates demand for the call centers that India has become famous for. Fully 69% of the jobs in India are at risk of displacement due to the emerging revolution in artificial intelligence and automation. India’s IT industry which is based upon a strong and perhaps excessive commitment to the study of highly technical subjects in its schools and universities needs to adapt to a new era in which technical acumen must be balanced with the creative insights and empathy that flow from the study of the humanities. It is the latter skillset that empowers one to climb the value chain to become a software architect rather than a tester; a systems designer rather than a low-end coder; a creator of content rather than a grunt IT worker.

KrishnaK · Post by **KrishnaK** » 17 Jun 2017 00:30

KJo wrote:I think Harpreet Singh is overreacting.
Indian IT's business model is to pick whatever is "hot" in the market, train the vast armies of graduates in that technology and send them to the US and market them with a "buy 1 get 10 free" model. This has worked very well for years. I don't see why it won't work now.

In 1999 it was Java/C++ and Y2K. In 2017 it will be Artificial Intelligence and Machine Learning and Analytics.
Teach Machine Learning and Data Science to 21 year olds in a 2 month crash course, and exploit demand.

It's harder to teach ML/AI compared to Java/C++/Cobol.

SRoy · Post by **SRoy** » 17 Jun 2017 13:16

That article is littered with expressions like "voice enable", "call center". According to MSM any living organism getting within 10 meter radius from a PC counts as a IT worker. So that article is bullshit.

So, according to the author, one is discouraged to be a low end coder, but needs to be creator of content. I guess Harpreet doesn't even know where his elbow ends and ass starts.

That bullshit article apart, the picture is complex.

I just interviewed a tester with 4 years of working experience from TCS. Absolutely appalling. Disgusting. The candidate couldn't even answer a single question straight. I had a similar experience with project manager candidate from TCS some years ago. Same sub-human IQ material. In my opinion TCS is a dodgy organization picking up projects employing underhand means, otherwise I cannot explain presence such quality of human resource material. Future is bleak for these people AI or no AI. I could be wrong, because TCS might then be bribing up the next suckers down the line to keep revenue flows open.

On the topic, I have a word for those taking AI lightly and those hyping it up.

Those taking AI lightly should understand, you simply cannot train armies of people to a knowledge area that requires you to posses a certain minimum intellectual level. Most doesn't have it.

Those hyping AI should note that "harder" is a relative word. Java/C++/Cobol in a single sentence also means superficial knowledge of legacy and current programming languages. Sorry to point out that. The top cream of these programming language practitioners must already be practicing ML/AI by themselves.

Please also note that there are many systems that are capable of modifying their run time behavior solely on based of changing inputs and they do undertake certain degree of decision making themselves. I have seen those in Telecom and Transportation. Of course, one must add that the designers themselves may not have thought of a formal structure or established algorithms to design those system. Or maybe they did. But they exist. Built by average people.

Indian IT i.e. IT proper consisting of SW product development, bespoke software development, software maintenance, system integration and testing needs to get rid of the whole generation that got in during Y2K rush.

prahaar · Post by **prahaar** » 17 Jun 2017 13:33

KrishnaK wrote:
KJo wrote:I think Harpreet Singh is overreacting.
Indian IT's business model is to pick whatever is "hot" in the market, train the vast armies of graduates in that technology and send them to the US and market them with a "buy 1 get 10 free" model. This has worked very well for years. I don't see why it won't work now.

In 1999 it was Java/C++ and Y2K. In 2017 it will be Artificial Intelligence and Machine Learning and Analytics.
Teach Machine Learning and Data Science to 21 year olds in a 2 month crash course, and exploit demand.
It's harder to teach ML/AI compared to Java/C++/Cobol.

A few ML experts could leverage a bunch of software engineers to implement modifications and run training/etc.

Tools for generating various optimal datasets. There is a large body of work which goes into creating production systems to leverage ML, which can be done by non-DL-algorithm experts.

Singha · Post by **Singha** » 17 Jun 2017 17:08

I would like to have a clear definition of AI. Negi always talks of "rule engines" as a layer below true AI. I 100% know the MSM has no idea of what ML/AI is but let us ourself be clear.

is "rule engine" a large bunch of rules of "if x do y" ..ie very precise if sensor reports any value in range 0.1 to 0.15 then increase temp by 0.1C
are the ubiquitous chatbots based on rule engines or they do NLP and hence fall into rule-engineMKI / AI-lite ?

and true AI is a system that does things not precisely defined by the creator of the system but based on some "loose guidelines" or "learnings from M past inputs"(ML) or even averaging a moving avg working value over 100 past inputs and creator does not know what these values will be ?

would the tcp congestion avoidance mechanisms be called AI? it is proven, solid, used in every phone and computer and works really well. truly a genius work of "peasant math" that anyone can understand if they read it up, than some obscure math like "proving fermats theorem" which needs math Phd background to start getting into

https://en.wikipedia.org/wiki/TCP_congestion_control

ManSingh · Post by **ManSingh** » 19 Jun 2017 05:05

Question to all experienced IT bros.

How to switch career streams mid-way?
For ex: If I have an embedded systems background( 8+ years ) and want to move to a different field ( say deep-learning/computer vision ), how do I do it?
I do get the learning curve, which I do find manageable. It is the past experience demanded by employers which is tough to overcome. For ex: firms asking for 3+ years experience for theano and 8+ for machine learning. In reality theano in it's present form was launched only in 2014 and the current wave of machine learning hasn't been there for 8+ years? In any case, I have zero past experience.

Any tips welcome.

SRoy · Post by **SRoy** » 19 Jun 2017 09:30

Move to a company that has both the streams of business. Try for inter departmental transfer after that. This naturally happens whether we like or not, recruited for some work and then shifted to somewhere else.

This process is hundreds of times easier than just trying to switch via open market opportunities.

Sometimes companies do recruit people from other domains, but there the criteria usually is very strong skills in one technology area. Worked for me.

Public profile that show cases that skills (e.g. portfolio of work) is a good way to get noticed. This would require investment in terms of time and hard work outside of regular working hours.

ArjunPandit · Post by **ArjunPandit** » 19 Jun 2017 10:08

ManSingh wrote:Question to all experienced IT bros.

How to switch career streams mid-way?
For ex: If I have an embedded systems background( 8+ years ) and want to move to a different field ( say deep-learning/computer vision ), how do I do it?
I do get the learning curve, which I do find manageable. It is the past experience demanded by employers which is tough to overcome. For ex: firms asking for 3+ years experience for theano and 8+ for machine learning. In reality theano in it's present form was launched only in 2014 and the current wave of machine learning hasn't been there for 8+ years? In any case, I have zero past experience.

Any tips welcome.

Not in IT so cant comment from that perspective, but doing some MOOCs/certifications also helps. Some skills e.g. scripting e.g. PERL/python are transferrable.
Three key aspects to Big Data/ML/AI that i have seen FS industry
1. Data/Infrastructure: Capabilties, e.g., Hadoop/Hive etc.s implementation at enterprise level; capability to use unstructured data, e.g., clickstream
2. ML techniques: RandomForest, GBM, SVM, Skytree, Ensemble tools etc.
3. Business usage of above two: in high processing matches, merges, reccomendation engines, fraud detection, prospecting/underwriting decisions etc.
**************************
Added later on;
I do hear talks of Robotics, NLP etc in AI, but nothing tangible in FS industry that I have seen.

Banks and FIs are still in exploration/investigation mode

Singha · Post by **Singha** » 19 Jun 2017 10:54

nobody has answered my q above, but imo there is no need to in lungi shiver mode over AI/ML if all you want is to "use" it rather than write libraries - which will be dominated by cash rich or specialized cos.

as in any technology let us say compiler, the core tool and libs are made by a few people (eg gcc, visual c++) while a vast number of people make use of applications on top of this. likewise i expect the current wild west hype to settle down in terms of what works where , better trainings, better books and tools to use the underlying AI/ML features . people should be good in programming and general logic as always. if "using" AI/ML needs Phds in maths and stats background to "get in" , it will never dominate like RDBMS or mongodb/cassandra or tcp/ip due to barrier to entry.

in a few yrs it will be no different than learning a new database pkg or a new library in a language and will be part of life. infact it already is - if we are using google, fbook, chatbots to mediate our interaction with the world we are already using ai. banks might already be using ai for flagging fraud and black money.

I dont think writing AI/ML/NLP code will be day to day activity for 99.99% of programmers. but using libraries may be for some.

SRoy · Post by **SRoy** » 19 Jun 2017 11:28

@Singha,

There is nothing to answer. The only people that are in lungi shiver or shalwar soiling mode are the one that get worked up to every hype that come around in 4-5 year cycles.

Your bot example is a good one. That's what many people are working on minus the hype.

Singha · Post by **Singha** » 19 Jun 2017 11:51

yes I agree, people should overall keep their programming, logic and data structure skills upto the mark ..and spend some time reading about new stuff... and not be dogmatic about one language or framework because these will change for sure every 5 yrs. C/C++ being so long lived is a outlier. perl died when python came and now Golang has also emerged claiming to be even more concise.

i have heard there is a "coding round" even for director recruitments in SV. would love for the same to be applied on all managers and directors here.

netz is btw toying with a proposal for everyone to have a programming test every quarter. heard it from a higher up - dont know if he excuses himself and his tribe of managers from that.

there is no comfort zone and cushy glide path anymore once we are fairly senior....every week is a fight to add some value ... like bheesma and drona we will need to keep fighting and adapting every day to survive in the battlespace. these folks never mentally or physically 'retired' and could lay waste entire armies on their own. hands on and honourable till their last day.

gauravsh · Post by **gauravsh** » 19 Jun 2017 13:06

Singha wrote: netz is btw toying with a proposal for everyone to have a programming test every quarter.

Interviewed lot of people from Samsung last 4-5 months. All were very well prepared in DS fundamentals. But if you scratch or tweak things a bit and ask them to write code they would falter. Finally i did ask one of the candidates what was going on. It seems Samsung has started some in internal qualification for programming fundamentals. They have trainers for that. Employees are expected to clear a minimum level. Most of them used this to prepare for interviews anyhow

sum · Post by **sum** » 19 Jun 2017 13:54

It seems Samsung has started some in internal qualification for programming fundamentals

They are pretty anal on that and use it a lot to keep the pink slips rolling

Singha · Post by **Singha** » 19 Jun 2017 16:19

hmm i also intviewed one from samsung noida over video. he wrote good code so my vote was a yes.

NRao · Post by **NRao** » 19 Jun 2017 17:24

Singha wrote:I would like to have a clear definition of AI. Negi always talks of "rule engines" as a layer below true AI. I 100% know the MSM has no idea of what ML/AI is but let us ourself be clear.

{Folloing up on your next post}

Is there a reason you are not satisfied with the wiki definition? Trying to understand.

any device that perceives its environment and takes actions that maximize its chance of success at some goal

What is inherent in that def is the concept of "learning" - as part of "takes actions".

NRao · Post by **NRao** » 19 Jun 2017 17:37

ManSingh wrote:Question to all experienced IT bros.

How to switch career streams mid-way?
For ex: If I have an embedded systems background( 8+ years ) and want to move to a different field ( say deep-learning/computer vision ), how do I do it?
I do get the learning curve, which I do find manageable. It is the past experience demanded by employers which is tough to overcome. For ex: firms asking for 3+ years experience for theano and 8+ for machine learning. In reality theano in it's present form was launched only in 2014 and the current wave of machine learning hasn't been there for 8+ years? In any case, I have zero past experience.

Any tips welcome.

I have found that pretty much everything falls into two buckets:

1) Where you have an expertise and someone else contracts with you to use your expertise. They provide the details (design) and you complete teh task (code). In this model you are the SME in your own field, but not in the field you are contracted to support. Typical coding projects. Someone provides the designs and you code. Done (simplifying here)

2) Where you are the SME in your field, BUT you need some knowledge of the another field which is using your expertise. SAP. Architecture, code, etc, etc, etc, but there is a need to know accounting or ERP or finance. You cannot just be happy with just coding

AI/ML/whatever typically falls under the latter. In fact, IF you want to be an "expert", you better know a whole lot about the "other" field. So, neural networks? You better be pretty good at how the human brain operates. NLU, you better be great at gramer of various languages. Pattern recognition, you better be good at what are patterns in the domain of interest. In fact, Big Data, I suggest you know more about statistics that computer science (you can always hire or read up on Hadoop, but not the other way around without investing years in stats). Robotics is a great deal about motors and previsions and lubricants and ................ In all such acses "code" is important, but there are other factors that could/may, invariably do, take over the landscape.

And, typically in AI, the "other" fields are related to human anatomy, nervous system, brain, eyes, ears, etc. And that is where most of us fumble and give up.

In short, be prepared to face great surprises and have the fortitude to overcome them.

NRao · Post by **NRao** » 19 Jun 2017 23:51

OK. I happen to stumble across a TED talk from 2015. A very good example of ML, as applied to the "field" of basketball.

https://www.ted.com/talks/rajiv_maheswa ... s#t-500145

Please watch the entire vid, but pay attention at 8:30.

Please watch it and then decide if you *really* want to get into something like ML (at THIS level) (you could just be a plain programmer too).

And, to be clear, THAT talk was from 2015. So, they have been at it for some 5+ years, so I bet they go back to at least 2010, if not earlier.

Which brings me to another dimension of AI - time.

ManSingh · Post by **ManSingh** » 20 Jun 2017 06:10

NRao wrote:
ManSingh wrote:Question to all experienced IT bros.

How to switch career streams mid-way?
For ex: If I have an embedded systems background( 8+ years ) and want to move to a different field ( say deep-learning/computer vision ), how do I do it?
I do get the learning curve, which I do find manageable. It is the past experience demanded by employers which is tough to overcome. For ex: firms asking for 3+ years experience for theano and 8+ for machine learning. In reality theano in it's present form was launched only in 2014 and the current wave of machine learning hasn't been there for 8+ years? In any case, I have zero past experience.

Any tips welcome.
I have found that pretty much everything falls into two buckets:

1) Where you have an expertise and someone else contracts with you to use your expertise. They provide the details (design) and you complete teh task (code). In this model you are the SME in your own field, but not in the field you are contracted to support. Typical coding projects. Someone provides the designs and you code. Done (simplifying here)

2) Where you are the SME in your field, BUT you need some knowledge of the another field which is using your expertise. SAP. Architecture, code, etc, etc, etc, but there is a need to know accounting or ERP or finance. You cannot just be happy with just coding

AI/ML/whatever typically falls under the latter. In fact, IF you want to be an "expert", you better know a whole lot about the "other" field. So, neural networks? You better be pretty good at how the human brain operates. NLU, you better be great at gramer of various languages. Pattern recognition, you better be good at what are patterns in the domain of interest. In fact, Big Data, I suggest you know more about statistics that computer science (you can always hire or read up on Hadoop, but not the other way around without investing years in stats). Robotics is a great deal about motors and previsions and lubricants and ................ In all such acses "code" is important, but there are other factors that could/may, invariably do, take over the landscape.

And, typically in AI, the "other" fields are related to human anatomy, nervous system, brain, eyes, ears, etc. And that is where most of us fumble and give up.

In short, be prepared to face great surprises and have the fortitude to overcome them.

Thanks NRao and everyone else who replied.

NRao, I perfectly understand what you are saying. Coding alone definitely can not make one happy.

I might be totally wrong but my view is that I can still get into ML, DL and AI even without knowledge of statistics, maths, neural networks etc. I don't intend to implement machine learning algorithms or do a research Phd. I am going to use simple packages and already established algorithms.

My view is simple: Python -> ML -> DL -> AI. I got it from here: https://www.superdatascience.com/course-map/.

These guys have done a great job. All their courses are available on udemy.

I intend to follow the right most path. In fact, I already started.

Also you are right about applying ML etc to my current domain( embedded ). I don't intend to become a data scientist.

In fact this is the actual reason I intend to study ML in the first place. Many current jobs in embedded domain( embedded doesn't always mean low power ) actually mention "Knowledge of machine learning is a plus".
Currently such jobs are a trickle and frankly not worth the effort. But who knows a decade down the line. I might be completely offtrack though.

vivek_v · Post by **vivek_v** » 20 Jun 2017 09:13

Singha wrote: .........................................................................
would the tcp congestion avoidance mechanisms be called AI?
..............................................................

I would not call it AI. Also, I am not sure what rules engines are being used with either Machine learning or Deep learning since those are not really mentioned in this thread. If that is clarified then it would be a lot easier to get an idea of what is being talked about.

To clarify, the ML/DL techniques are basically "Universal Approximators" (i.e) these techniques can approximate any function. The example you have given is an example of a deterministic system where the mathematic rules seem to be clearly defined.

For example, you can take an FFT directly (or) if needed you can train a deep neural network to approximate (learn) for the same. But this will not have anything close to an efficiency or even accuracy of directly applying the math rules and writing code for the same. Actually, in a panel discussion, someone was mentioning on why FFT itself could be considered as an example of a Deep neural net but that's OT for this thread.

Now if you have linearly separable data (or) you if you know your domain knowledge is good enough, you can extract features to make the function linearly separable and train a linear classifier like Logistic regression or Linear SVM and model a function for your problem. People also used to employ something called as a "kernel trick" with SVM to separate non-linear features. This was not something new and people been using these stuff in some form for the past two decades or so but there was a fundamental limitation to how much feature engineering you can do on data like Images, Audio, Signals or even Text.

The real breakthrough came in 2012 when neural networks made a comeback in form of paper from Alex Krizhevsky, Ilya Sutskever and GeoffreyHinton (Alexnet paper) with beat the previous records achieved by feature engineering and linear SVM by a huge margin and they did not do any form of "feature engineering". This is father of current revolution in AI so to speak. The term Deep learning was coined to make it sound "cool" since it would be too much pain to explain why Neural network works now.

Coming back to your question on the TCP congestion case, why would you want to approximate a function (or an approximation function) where the real function is known?

NRao wrote:AI/ML/whatever typically falls under the latter. In fact, IF you want to be an "expert", you better know a whole lot about the "other" field. So, neural networks? You better be pretty good at how the human brain operates.

Why would anyone want to know how the brain works for understanding Neural Networks. They have absolutely practical relation since for one dendrite performs nonlinear computations, Synapses do not have a single weight like software neural networks but are non-linear dynamic systems, there are multiple types of neurons...etc.

All we do in Neural Networks/Deep learning is to inject some form of nonlinearity using neurons (Tanh, RELU, Leaky Relu, Maxout..etc) so that gradient descent would converge to an optimum during backpropagation. If we don't inject this non-linearity then the whole system could be simplified to linear regression which is not what we want.

negi · Post by **negi** » 20 Jun 2017 17:39

Singha wrote:I would like to have a clear definition of AI. Negi always talks of "rule engines" as a layer below true AI. I 100% know the MSM has no idea of what ML/AI is but let us ourself be clear.

is "rule engine" a large bunch of rules of "if x do y" ..ie very precise if sensor reports any value in range 0.1 to 0.15 then increase temp by 0.1C
are the ubiquitous chatbots based on rule engines or they do NLP and hence fall into rule-engineMKI / AI-lite ?

Let us take example of chatbots ; the issue at hand here is if I pay say 50k USD per anum to a tech support person to take calls from customers I would have to an army of them depending on the size of the business . However the challenge here is even a sophisticated NLP model requires supervision or human training (to be precise fed with carefully crafted data which captures all the uses cases) now for a business the question is it worth it to invest into a state of the art chatbot that can handle these questions like a human would ? Generally mature organizations realize it pretty early that it is better to use chatbots as first level of customer service and still retain the humans for questions or interactions which are still very very hard to model for . I mean if a customer has poor language skills simply because English is not his mother tongue then chances are chatbot might not even be able to parse right keywords from the question , responses generated to further clarify the question might get responses which are again poorly structured . Another fundamental problem is human nature i.e. we all prefer to talk rather than type if the interaction is going to be any more than 2-5 minutes and bots that can answer phone calls need to be even more sophisticated because now they have to be able to correct phonetic errors, mother tongue influence and pronunciation issues among many other variables.

Long story short since the effort, time and money needed to build a sophisticated chat bot might not lead to resolution of all kinds of customer queries companies tend to employ them for kind of interactions which are a bit more closely defined and rigid in terms of where the conversation can go.

IBM Watson program that won Jeopardy is a classic example of both how AI has advanced and yet how much more is left to be done ; Jeopardy has a fixed format and a pattern as to how questions are asked so I am sure all this was part of the Watson model . Now that there is news about Watson being used by nurses for decision making or getting recommendations I am sure that the implementation team would have to tweak the model for that kind of interaction and then follow it up with loading of prepared data set for training once done and plugged to all the data sources for relevant domain it would be ready for use.

I forget the name of the company which Alphabet acquired based out of UK which claims to make AI systems which can akin to humans observe a computer video game and start playing it without human training .

NRao · Post by **NRao** » 20 Jun 2017 18:53

I might be totally wrong but my view is that I can still get into ML, DL and AI even without knowledge of statistics, maths, neural networks etc. I don't intend to implement machine learning algorithms or do a research Phd. I am going to use simple packages and already established algorithms.

Why would anyone want to know how the brain works for understanding Neural Networks. They have absolutely practical relation since for one dendrite performs nonlinear computations, Synapses do not have a single weight like software neural networks but are non-linear dynamic systems, there are multiple types of neurons...etc.

Sure, one can use a packaged product (IF I am reading the posts right that is) to provide results for any client.

However, even within a packaged product, IF one were to understand the correlation between the real-world mechanics and the package, one can tweak/configure the package to provide far better results.

So bail gaddi use case: A person who knows lubricants well > a person who knows he needs to lubricate the axle > one who just rides the cart (without being aware that the axle needs basic maintenance).

From experience I can state that many a times, even a small differentiator makes a diff. As an example I have been selected (as a singleton consultant) just because I claim I know how to code for clustered environments - sequential processing in clusters. Simple stuff, but that has got me into doors, beating out OEMs. Get better rates too. Very well traveled - SA to NZ.

The other factor is about sitting at the table. Do you want to be at that table? Be part of the team that decides? Or do you want to be someone who makes good money (which is fine) and is told what to do? IF it is the prior, then you have to provide something to differentiate. And, coursera, etc - great as they are (I use them constantly) - but they only make you part of the herd. I use coursera, etc to figure out what everyone knows and then figure out what I need to do to keep ahead of that herd. And then let the "fame" precede you - no need to write resumes.

Singha · Post by **Singha** » 20 Jun 2017 20:54

only 0.1% of engineers have the name needed not to write a resume. like the inventor of java recently joined amazon. there would be 100s of accomplished and up to date engineers like him but without the name and fame.

Singha · Post by **Singha** » 20 Jun 2017 21:28

https://thenextweb.com/artificial-intel ... w_zXJbTexa