Innovative intellectual odd, one to "AI-enabled" as the theme of the company, is a subsidiary of Innovation Works, was founded in March 2018. Do not know how to level, to inform the interview went.
job requirements:
Image vision algorithms intern
250-280 / day Master of Beijing four days / week internship 7 months
Job Responsibilities:
1. responsible for the development of computer vision algorithms related issues involved include, but are not limited to: detection, segmentation, classification;
2 tone model parameters responsible for deep learning and performance optimization, applications implemented on an industrial deep learning image;
Requirements:
1. Master's degree or above, computer, automation, mathematics and other related professionals;
good 2. English, have good English reading and literature search capability;
3. Computer expertise in algorithm design, data structures, etc. to master good;
4. depth understanding of learning methods and image correlation applications, using at least one depth skilled learning tool, such as TensorFlow, Pytorch, Mxnet like;
5. master C / C ++, Python in languages such as familiar Linux operating system;
6. image priority areas related papers published experience;
7 days a week for at least four days, practice for at least three months.
Content of the interview:
Practice self-introduction (simply speaking with the origin of the image, complete set of undergraduate, graduate programs balabala ......).
Let me introduce two projects related to the GAN (I'm talking about half of the project described in the context, and then talked about my job is what, innovation is what). Did not ask the details, did not ask convolution layer BN layer ResNet layer image processing, he asked me which model is used (cycleGAN).
To say I made the papers did not write the name just write a journal (I said was received but not yet see Journal).
I asked what the traditional image processing methods (gray binary expansion and contraction of edge detection filter ...... (written on a resume ah brother)).
Finally I saw recently participated in the contest Tencent algorithm, the processed data 6G, just super happy, I would say he kinda need these.
Then test a few super simple arithmetic problem:
- n steps, may be on a grade 1 or 2, Q There are several methods on the step: f (n) = f (n-1) + f (n-2)
- Given m, n, returns m ** n.
- There are all kinds of files under a folder, achieve a picture taken from multiple random (with replacement problem), so that has been read into memory fetch from memory, did not read into memory enrolled into memory. (Image files to establish a list of randomly selected images to determine whether the picture in it read in memory from memory, not just read from the disk, you can use a dictionary to store in memory of the picture, the dictionary has to read from the dictionary, not just added to the dictionary from the file reads)
- [There is also a problem, forget, are not difficult]
Very pleasant to talk over, he asked me what the problem. I asked here is mainly what kind of project, he said, quality testing machine when there is a problem of unbalanced data sets, and more qualified, unqualified and less. His job is to generate some sample data set failed to improve the quality inspection identification accuracy.
It ended quickly and happily, go to HR little sister received a letter of micro friend request. He said the daily wage of 280, need to confirm my willingness to work (I am of course very willing friends ha ha ha), and finally told me to finish the examination and approval can send offer, happy.
to sum up:
Scrape together only say that a coincidence too lucky, I just research papers and overlap. The project did not ask extremely fine, estimate the newly established interviewing many people find so difficult counterparts. Aspects of the algorithm, better brush up title, hard work pays off, although not large brush to the level of God, but the fact that these days there is no question brush efforts white. I feel a simple test, only that I have learned through the efforts, recall previous algorithm interview questions, very simple, but it was not super hard.
Come on, it opened a new chapter!