Where are the opportunities for machine learning startups?

http://www.vccafe.com/2015/12/01/where-are-the-opportunities-for-machine-learning-startups-guest-post/

by ColinWright • 11 years ago

46 19 11 years ago

vdnkh• 11 years ago

I've been getting my linkedin spammed by recruiters for a machine learning startup. I ignore them all (despite actively looking for a job). Why? It's a pie-in-the-sky idea which seems more interested in generating hype than actually delivering results. To me, the product sounds too good to be true, and is trying to replace a service which many people still do not trust humans to do. The technology "is just no there yet" for a lot of the claims proposed - and I doubt that a startup focused on making money can accomplish more than academic institutions focused on research.

evandev2• 11 years ago

I work for a startup[0] that does quality control through machine learning for the beverage industry. We're focused most on beer, followed by coffee and spirits.

[0] https://gastrograph.com

andr3w3212• 11 years ago

Where do you find these ML wizard conusltants? I've been looking to hire someone part time for a project to help build a model but most good people have no time and I don't trust that anyone on elance or similar is any good.

nalourie• 11 years ago

Depending on the project / time line I'd be interested to help. Is there a way to get in touch?

andr3w3212• 11 years ago

Whoops on phone and replied to wrong comment. If anyone is interested contact me andr3w321@gmail.com

davmre• 11 years ago

Email older PhD students at your local university, assuming you have a strong local university. Describe your problem in a few sentences to get them interested, and offer competitive compensation ($100-$200/hr depending on location) in exchange for a few hours per week of their time.

JasonCEC4• 11 years ago

The short answer is: _you don't_. That's not how machine learning works at scale.

To quote from a great recent article: (I can't find the link, sorry)

It's one thing to create an excellent fraud detection model in R, and quite another to build:

  - Fault-tolerant ingest of live data at scale that could represent fraudulent actions
  - Real-time computation of features based on the data stream
  - Serialization, versioning and management of a fraud detection model
  - Real-time prediction of fraud based on computed features at scale
  - Learning over all historical data
  - Incremental update of the production model in near-real-time
  - Monitoring, testing, productionization of all of the above

You don't build a data team out of a single person and tack on an easy model to build a company - it takes a team to build a real data project, and those teams are hard to find, hard to recruit, and hard to make successful.

evandev2• 11 years ago

We're based near a large University and hire Masters/PhD students about to graduate first as a part time/intern then full-time. It has worked out pretty well for us.

JasonCEC4• 11 years ago

Evan beat me to it :)

Evan is our Lead Engineer, and I'm the CEO.

We believe vertical integrated SaaS focused on prediction will reshape a lot of industries.

exelius4• 11 years ago

I believe that too. But I also believe that large ERP vendors will be the ones delivering this value, because integrating predictive modeling into a company's operating model is the hard part. They're also a lot more vertically integrated, plus their products work with your existing ERP / forecasting system.

Though it does seem like you guys have found a niche - quality automation in general is booming right now, and ML predictive models can be very helpful in certain industries like food and beverage. I'm guessing you guys probably work with smaller/younger companies that don't have the kinds of problems your Kraft/P&G/J&J do - there's certainly plenty of opportunity there.

"Prediction" in general can be hard - the biggest problem can be getting good data in the first place, and fixing a company's data gathering problems is out of scope of any sane software shop. But it's right in the wheelhouse of an IT consulting shop, which is where I'm seeing a lot of ML guys end up.

exelius4• 11 years ago

This article skirts around the issue, but the biggest (and most common) mistake I've seen machine learning startups make is starting with an algorithm and trying to productize it. When you do this, you end up with a solution looking for a problem, which is never a recipe for a successful startup. So in addition to finding a problem that fits your solution, you then also have to build a product around it. And chances are, if your product is really great, you can build it without fancy machine learning models and it would be just as successful.

Machine learning, in my mind, is more of an enabler that makes great products more 'magical' - it's not really a product in itself. IMO it's a discipline that lends itself well to a consulting model - hire some ML wizards, unleash them along with a dev team on a problem for 6 months, then move on to the next one. This is mostly the "infiltrating the professions" - most consulting companies organize themselves around market verticals for the exact same reasons.

ThomPete• 11 years ago

Exactly!

Unless you sell machine learning algorithms of course :)

exelius4• 11 years ago

That's basically consulting :) There are very few truly novel applications of ML, and nearly all the major algorithms are open source. Using them together correctly and interpreting the output is the hard part -- which is perfect for scoping a consulting engagement.

nostrademons• 11 years ago

Also, the critical element for machine-learning models is training & testing data. Many startups lack access to large quantities of data; since data is proprietary and often expensive, it's really a competitive advantage for large established companies, not startups. If your startup is focused around a machine-learning model, your first priority should be to acquire lots of data - and usually, if you can acquire lots of data, you have a valuable product in its own right without the machine-learning.

Totally agree about ML being an enabler to make great products better, not to bring new products into existence.

tiredwired• 11 years ago

A machine learning startup should use their machine to answer this question.

dafrankenstein2• 11 years ago

One idea is to use machine learning for training drones planes and quadcopters....that will make a startup different than others! Actually applying ML in a product will stand out in the market if the job is done properly.

Where are the opportunities for machine learning startups?

Related Stories