Perguntas de entrevista de vagas de Machine Learning Developer da empresa SAP

I was asked to provide basic information such as visa status, notice period and expected salary after submitting my online application. A mini-project with problem description and data was sent to me and the time to complete it was one week. The requirements were similar to Kaggle competitions and a notebook should be submitted as the final report. After clearing the challenge with accuracy of 80% using CNN model, I was invited to the face-to-face interview through a phone call from HR of SAP in Philippines. I arrived 10 minutes earlier than the agreed time for interview. Unfortunately, SAP has two different office in Singapore and HR gave me the wrong address for interview in email. The given address is for SAP reception desk in one building but floor-room no. is for SAP IT department at the other building and there is no short-cut connecting these two buildings. Even the property manager of the building was confused by the address I received. It took me about 30 minutes under tropic sunlight to get to the correct office for interview after I found the SAP reception desk and got the correct address. The two round of interview was conducted in a small meeting room and there was some lunch leftover on the table (I remembered this because my elbow touched it accidentally although I tried to avoid so during the interview ). This gave me unprofessional first impression about the interview. Two interviewers of each round were from the same development group which means I had to introduce myself and my projects twice for two different groups. Though small, the meeting room was big enough for 6 person and I don‘t know why the four interviewers cannot come together. One of the interviewer of the first round was impolite and unprofessional. He interrupted me by saying that he “can not understand my research at all” while I was introducing my previous experience as asked. He even asked me about "how to choose one out of k models in k-fold cross-validation". I answered to him with the definition and implementation of CV and added that "CV is used to evaluate the capability to predict unseen data and avoid overfitting for ONE model. We dont have k models but one model with k sets of trained parameters". The interviewer apparently misunderstood the model/method and the instance of model/method in the context of CV. One can compare a SVM and CNN model for a specific problem on training data with k-fold CV (k-1 fold for training and 1 fold for validation). But once CNN is proved by CV to be better than SVM, one should use ALL the training data to train the CNN model. Simply, CV is for model checking but not for model building. I can feel that his attitude began to become impatient after this question. He then asked some questions (such as "why not implement SVM using CNN", "why use RNN rather than CNN to handle time series") that aim at neither to obtain an answer nor to assess an interviewee's ability of problem-solving but only to prove that "I am able to attack your opinion immediately after you raise one". I explained with some corner cases for these questions and also pointed out that it should be avoided for most of the time because we must select the right model to do the right thing. The other interviewer with a doctoral degree of the first round was polite and asked me about details of my machine learning projects and also answered my questions about his group. The second round of interview was just as normal as other SDE interviews with standard questions.

Asked about the projects on your resume. Then asked some basic ML questions, such as purpose of PCA and saddle points in training neural networks. Finally, a code test on a white board.