I am a computer science doctoral student working on machine learning (ML) research. Meanwhile, a startup approached me to consult/advise them in areas related to algorithms in ML. Needless to say a lot of work that I do as part of my research encompasses ML deeply. They want me to sign a mutual non disclosure agreement (NDA)/ non compete clause before i engage as an independent consultant with them, while pursuing my program full-time. In this context, they know well that I will be pursuing my program for at least the next 3 years.

I have a few problems here.

  1. The confidentiality information covers:

    ...techniques, know-how, processes, algorithms... related to current and future proposed products and services... and includes, without limitation, their respective information concerning research, experimental work, development, design details and specifications, engineering..."

    I am worried about this part mostly. Most ML research uses some form of classification or clustering in various domains. Every other startup is mining some patterns that they may use to their advantage. The methods used are generally known in the ML community and published in existing literature. Can a startup claim that an "algorithm" (which already exists in the literature) was used by me in some form at a later point of time for work unrelated to that domain? Additionally, I intend to use or publish research methods developed as part of my doctoral research during its course or in future. Does "without limitation ...respective information concerning research, experimental work, development" restrict me from doing this i.e can the organization restrict me from publishing work done in my doctoral program by claiming that I was also working with them during this period, so they have a right to what i publish during this time?

  2. Regarding confidential information:

    "...if orally disclosed, designated as confidential at the time of disclosure and marked in writing as “confidential”."

    Does this mean that anything said orally related to ML is considered confidential?

  3. Non-Compete clause:

    "...with a direct competitor of Company while engaged with Company and for thirty-six months following the termination or completion of this engagement..."

    As an ML researcher, I will be working on ML in the future years to come. Asking me to stop working in ML for 36 months is equivalent to asking "don't work for 3 years". Additionally, the startup does work which is similar to the kind of work a lot of analytics/data science related firms do. Most of ML research broadly falls into pattern mining in some form which almost all organizations engage in. In this context, does 36 months not seem a bit unreasonable?

  4. What exactly can i get them to write in the NDA which informs that they are aware of my research position and do not expect IP for my academic research?


You are dealing with a legal document, and you are asking questions about legal language. The only way to get answers to your questions is to consult a legal expert. Any advice you receive here will necessarily be inferior to that dispensed by the legal expert.

Having said that, you're asking good questions, and during the negotiation part of the discussion you can and should discuss your concerns with a legal expert and with the company representatives. Negotiations are a normal part of any contract discussion. The document you have was written by their lawyers to provide his client—the company—with maximum legal protection, in their best interests. The job of your lawyer is to provide yourself with the same protection, in your best interests.

If you are concerned about the cost of hiring a lawyer, do the math; the total amount you expect to receive from the consulting should be significantly more than the cost of the lawyer, and the potential cost to you of not hiring a lawyer may be pretty large.

Regarding finding one, speak with the faculty in your department; chances are at least one of them has consulted in the past and can provide you with both advice and hopefully the name of a qualified lawyer.

  • 3
    +1 for consulting a lawyer in your case. IANAL, but I have some background in IP law. The agreement you are considering between you and the company may well have serious ramifications for your future activities and career. Yes, lawyers cost money. As eykanal said, they've got their lawyers to protect their IP. You need to be certain that any agreement you enter into will not constrain your future activities adversely. Find an IP lawyer, preferably one with experience in dealing with computer software. – Nicholas May 24 '13 at 9:37
  • 1
    I would add that 1) you must negotiate that 36 month no compete down to 12 months or less, it is unreasonable, – daaxix Sep 19 '14 at 17:01
  • 1
    and 2) attempt to negotiate, written specifically in the contract that the assignment of IP to the company is strictly for specific machines developed by you, and the data, feature extraction, etc. specific to the problem for the company, and that all other general machine learning, and any algorithms or classification methods developed in the course of this work, are specifically excluded from the assignment of IP. You will need a lawyer to be precise about this. If you do not do 2), you will end up in big trouble with your thesis, with high (log) likelihood. – daaxix Sep 19 '14 at 17:01

Non-disclosure documents are almost always trying to squeeze the employee. Here are a few specifics that I can glean from the snippets that you've posted.

I'm not a lawyer, but honestly I don't think you need a lawyer to interpret much of the language in an NDA.


First, the NDA claiming that it covers processes and know-how relating ML methods refer to anything that you are working on related to current and future proposed products and services.

So let's say Company X is working on, let's say, movie recommendations and uses, for example, SVM (support vector machines) and collaborative filtering for the bulk of their recommendations. This does not mean that you can no longer talk about SVM and collaborative filtering with anyone - it DOES mean that you cannot talk about their particular implementation of SVM, you cannot tell people how they're doing collaborative filtering (ex: their features, their weights), you cannot mention what's in their databases and so forth. In fact, you probably can't tell people that the company uses SVM and collaborative filtering unless they tell you that you're allowed. If you develop some new technique of doing a movie recommender, you cannot tell people outside the company about it or publish it unless you get permission from the company. The key part of non-disclosure is that you basically need permission before you can talk about certain things, but common knowledge is common knowledge - it's not like the company can restrict you from talking about SVM and collaborative filtering. But if that's on their "super secrets" list of stuff, you better not go to a conference, give out your business card, and then say, "Well I know a lot about collaborative filtering and SVM".

Usually, "without limitation" DOES cover stuff that you work on outside of your work hours if it's directly relevant to the company. If there's a remote chance that your dissertation work and your publications might be inspired by, let's say, solving a company's problem or working with the company's data, then they actually do have the right to restrict you from disclosing your work without their permission, and may in fact claim copyright as well.

On the bright side, there are possible recourses for you. You can talk to them and ask them explicitly that you want to retain your copyright and that you want to reserve the right to publish about algorithms and so forth that you develop while you're there. You can promise to not disclose the company name and line of work. You can ensure that any papers you write go through their legal department so they can rubber-stamp it before delivery. These are things though that you want to ask about before you sign the NDA. Often, you CAN negotiate on terms like this, but it depends on their legal team.

TWO: Confidential

It seems to me that this is pretty clear. We've all taken formal logic, yes? That would mean that if something is orally disclosed to you, told to you that it's confidential, AND marked in writing as confidential, then you cannot talk about it. Unless they tell you, "Our implementation of SVM is confidential" and then point to the document that says, "The company's SVM implementation is confidential" then presumably you can talk about it.

That said, it seems like their lawyers made a mistake here, because you'd think it should be an 'OR' relationship here - "mentioned as confidential at time of disclosure OR mentioned in writing as confidential".

In any case, since SVM is basically public knowledge, most of what I mentioned under 1 would apply here. SVM and collaborative filtering and other stuff like that is public knowledge. However, don't say "Company X does SVM and collaborative filtering" in public.

THREE: Non-compete clauses

I can't really comment as well on this as I have no experience with non-compete clauses, but in some states they're not even legal (ex: California) and if they're phrased incorrectly to be unenforceable you might be off the hook.

I'd be more concerned about "who their main competitors are" rather than how long the agreement is. Such an agreement wouldn't prevent you from working in ML research, but it might prevent you from working at Netflix if your current company is in movie recommendations. Much of the time the non-complete clauses are made to prevent an employee from stealing corporate secrets and other implicit knowledge.

FOUR: Exceptions

Don't worry about where you'd put that stuff, you're not going to be writing the NDA.

You should bring all of your concerns with them though. The main thing is that you've brought it up with them and possibly written it down, maybe in an email, so that they can get their legal department to put it in the right location.

All of that said, I think your main concern is that "I can't talk about machine learning while working at this company." I'm 99% sure that isn't going to be the case. They don't care that you talk about general ML techniques or continue to do ML research, but will care if you're exposing corporate secrets, developing algorithms that might help their business with their data but not keeping it secret, or going to work for their competitors right after you finish with them.

Before hiring your own lawyer, I'd simply advise asking theirs about every clause in the NDA. That would save you some legal fees. If they are unable to answer, or if the answers aren't satisfactory, then you can consider getting your own lawyer to go through it.

  • 1
    This is a wonderful answer. The only thing I would add is that if you think anything might be interpreted in a way that could harm you, then add an addendum to the contract which states, in plain (not legal) English, what you believe you are signing (your statement about what you would and would not be able to do), and have the company sign that they agree with you. It doesn't hurt to have a lawyer review it but the law is much more accessible that many people think. – earthling May 21 '13 at 9:07
  • 2
    I think the main question is, even if you hire a lawyer to interpret a contract for you, what do you get out of it? You're out $100 and the lawyer will only tell you vague notes about what you can or can't talk (does the average lawyer know the difference between ML research at the University, and ML work at the company?) It's actually more productive to talk to the company, because presumably they WANT to hire you. The engineering team over there probably rolls their eyes at the NDA; their legal team at least has the power to change it! – Irwin May 21 '13 at 16:49
  • Yes, for ONE he really needs to specifically draft language stating that the only IP assigned to the company and subject to the NDA are specific features, weights, implementations of machines, and data, then have a lawyer review it for proper legal language. – daaxix Sep 19 '14 at 17:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.