Today is a sunny day, at the moment I do not know whether I am fishing, say I am fishing, I am going to share some dry goods, not fishing, I am sitting at the station in the code but not in the code.

In fact, I just feel that today, the weather is good, good mood, so I began my experience summary, also can be regarded as experience sharing (not to touch fish is not important), there is about “Machine learning Design” interview solution ideas.

The idea is to structure the expression:

Everyone who has experienced exam-oriented education should know that the test paper is graded according to the key points. If there is no expression logic, sometimes not only the manager does not know what you mean, but also they may not know what to express. Structured expression makes the answer appear logical. Structured expression, for a question, [Generally speaking, there are three answers, which are logically progressive, so that you will appear logical. If the answer is more than three, people will easily lose, so it is not necessary.]

  • For example, when asked a design question, the answer is complicated and has many parts.
This question can be considered in terms of two (numbers) : the user and the system.

There are 2(number) solutions on the user side

There are three (numerical) solutions on the system side

Because the interviewer does not know what you want to talk about, before you talk about details: explain the framework first, this will help the interviewer to follow your logic, even if you ask you carefully in the details, he will know that you have something to talk about later, will cooperate with you to finish.

  • There is also a common problem, most people do not know how to explain their own project experience (a small project, 1 minute to end the project will make the other party feel that the project is tough, a big project talk for 10 minutes, let the other party think, you did what project? Secondly, it will give the other party an impression of illogical thinking.)
Talk about your project: the topic should be relevant to the direction of the group, and the newer, better, better technology.

Suitable for friends who have done a lot of projects and have a lot of project choices.

  • Some students’ projects may be years of work and technology. How do you solve it?

At the end of the resume, give a Rethink, talk about two parts

First: reflect if do the same problem now, what can be improved (show your knowledge improved, not reflect on your shortcomings) do not rethink a lot of shortcomings, when the time comes, say I lied to you.

Here’s a list of things you can improve:

New technologies, such as < Logistic regression, can now use deep earning/LSTM, data size scale up, such as 10K data, now 100M, can use more complex models, system, Single PC -> Distributed system/GPU, training could be paralleled or in the cloud. Large companies have their own crowd sourcing judgment platform.Copy the code

Students may also be asked to talk about their own experience, and the interviewer will ask questions based on your experience.

What keywords and techniques do you highlight when you talk about your experience?

1. Relate to the other person

Second, new technology

Three, show your high level of content

Of course, if you are familiar with something, if you are asked something and you can’t answer it, it will be especially penalized. When it comes to which model to use, list two or three related models. The main point of the answer is to talk about their advantages and disadvantages, as well as the characteristics of the current problem. Based on these comparisons, you can decide which one to choose.

Proactively advance the conversation

Such as finishing a paragraph or answering a question: Does it make sense to you?

Before you talk about a complex system, say: I’m going to talk about a few components. If you feel interested in some particular ones, You can point out and we’ll discuss more over them.

Some of the questions are very vague, either on purpose or because the interviewer is poor at presentation.

This time

Rephrase + make concrete example

To actively probe the interviewer, remove the interviewer’s own errors, clarify requirements

Recognize your interviewer’s concern, and don’t wait for the interviewer to stop and start talking

Address the interviewer’s concern. Some interviewers are in conversation

Not every issue will be explicitly asked and then stopped for you to answer. When the interviewer is very unfriendly

To Break down the problem to solvable subtasks, we need to develop the principle of problem solving.

For example, when asked this question: “You deliver a ML product/system, but the accuracy of the reported system is much lower than that of your own test after users use it. What might be the problems? No log.”

From a problem solving perspective, the entire system has two elements: your ML system and the user’s application. Every element can go wrong. A good answer would be to break down the big problem into the two elements’ own problem + the problem of connecting the two elements.

The two elements’ own problems include:

Product problems, user application problems (use product domain and develop product domain inconsistent, customer data distribution and training data inconsistent, etc)

Problems of users (not using the system according to the designed way, measure’s method is wrong, using different metrics, etc from the developer).