Taro Logo

NLP Engineer Interview Experience - Montreal, Canada

June 1, 2020
Negative ExperienceNo Offer

Process

The process took approx 2 weeks. The first step is a standard hr interview and the second step is a technical interview.

The technical interview was the worst interview I have ever had in my career because 1- the interviewer was really unprofessional he was moving around during the interview making a lot of noise, he was also saying “efff” every 10 seconds and 2- He was really incompetent, his level of DL and NLP doesn’t level up to a beginner profile and I will explain this by examples of our conversation in the next section.

There was no question that I didn’t answer well and as mentioned before, the interviewer couldn’t even argue about something because his knowledge seems to be of a “learning by heart code” knowledge. Then after a bit less of a week they sent me the standard email of rejection without any explanation.

Looking at the interviewer’s LinkedIn profile after the interview, it turns out that the guy has worked there for barely 3 months. So Huawei makes newly arrived people interview candidates without any training and without any monitoring system as the interviewer called me with his personal cellular number. So basically the interviewer can say anything about the interviewee without necessarily being true. Lol ! This needs to be meditated!

Questions

He: what is IDCNN? Me: it is a CNN based architecture that is used in NER, it is based on dilated convolutions. He: why for sequence labelling use a CNN architecture? Me: because a CNN based architecture is parallelizable so the processing will be faster in contrast to RNN based architecture which is sequential. But I’ve seen in the littérature that generally RNN based+ CRF give better performance. He: why? Me: I’ve not seen formal explanations, all what I've seen is reported empirical results that show that better performance is given by RNN+ CRF. He: No because RNN catch the long sequence dependency. Me: the IDCNN I just explained catch also long term dependency He: how? Me: by dilated convolutions as I explained before. He: ok.

He: I am not understanding what is the difference between image segmentation and object detection? Me: object detection, you're trying to detect objects in an image by having a x,y,width, height and a label for detected objects whereas for image segmentation you want to segment the image as it is the world map catch the borders and give a label for each segment. He: ok. I am not saying that it is bad to ask a question but for a “researcher scientist” you need at least to be aware of tasks in machine and deep learning otherwise it doesn’t hurt to google them before the interview.

He: explain transformer Me: the architecture is an encoder- decoder, in the encoder there are two layers one for self attention and one a feedforward and after each there is a residual connexion + layer normalization. For the encoder, the input is the addition of the embedding of the tokens and their corresponding positional encodings. Then the interviewer interrupted me. He: what is the shape of the input? Me: every token has a shape of the embedding shape and in the original paper it is 512. He: No, it is batch size, sequence length, etc Me: do you want me to explain conceptually or the code ?!! He: ok.

He: why we can use lstm and cannot use transformer for long sequences? Me: With the architecture and formally speaking I am not seeing why you're saying that. Tell me the reason in your opinion. He: because in a transformer we need to fix the size of the sequence and in lstm we can have variable sequence length. Me: but in anyways we need to always fix the length at least batch wise for either by padding or truncating. He: No, because it is a tensor, In transformer it is fixed and in lstm in tensorflow we can vary the size dynamically. Me: as you said it is a tensor that you are giving as input, so for both and at least batch wise it needs to have the same length because it is by definition of the tensor and to alleviate the padding effect, in pytorch for eg you can use the pack padding sequence He: ok.

Was this helpful?

Interview Statistics

The following metrics were computed from 1 interview experience for the Huawei NLP Engineer role in Montreal, Canada.

Success Rate

0%
Pass Rate

Huawei's interview process for their NLP Engineer roles in Montreal, Canada is extremely selective, failing the vast majority of engineers.

Experience Rating

Positive0%
Neutral0%
Negative100%

Candidates reported having very negative feelings for Huawei's NLP Engineer interview process in Montreal, Canada.