注册一亩三分地论坛,查看更多干货!
您需要 登录 才可以下载或查看附件。没有帐号?注册账号 
x
最近作为new grad参加了Amazon的onsite, 因为博士生工作经验不多,尽力根据之前的实习经历和学校做过的projects想了一些高频的BQ题目的answer。希望对大家有一定的帮助!如果有用的话希望大家给添点米
我是按照STAR 方法回答的:
STAR method to construct the response:
-- Situation. Start off your response to the interview question by explaining what the ‘situation’ was and who was involved.
-- Task. Once you have detailed the situation, explain what the ‘task’ was, or what needed to be done.
-- Action. Now explain what ‘action’ you took, and what action others took. Also explain why you took this particular course of action.
-- Result. Explain to the panel what you would do differently if the same situation arose again. It is good to be reflective at the end of your responses. This demonstrates a level of maturity and it will also show the panel that you are willing to learn from every experience.
以下是个人的回答:
*1. Tell me about a time when you found a simple solution to a complex problem? (Invent and Simplify)*
I was in a meeting in my team where the visualizations and plots of different deep learning models for the project were discussed by the team members. Although the model visualization plots look intuitive, there are concerns that the plots are not so good due to the formats are not unified, and everyone needs to put extra time to write their own code to produce the visualization figures. After the meeting, I decided to try and find a solution to this problem. I checked the source code and met up with the individual team members, and it is apparent that they are using different ways to visualize the model outputs and are not familiar with other tools. After some investigation work, I write code to adopt a visualization API to our source code from an open-source tool and provide a standard for the plots. From that day on, the members could easily call the API to output unified visualization plots and do not need to write their own code. It makes the visualizations more intuitive and standard and saves them a lot of time.
*2. Describe a situation when you had to work with incomplete data or information? (Are Right, A Lot)*
I was working on a project during one of my previous internships, and the project required me to work closely with another member in other teams. We have been working very well together, and we have different skills and strengths which move the project forward nicely. However, towards the end of the project, my manager says that the team member has to leave for a while due to a home emergency, and I need to finish the rest on my own. The only way for me to finish the project was to use the limited information I had available, which was passed to me before he left. To make sure I could finish the project, I took lots of research to figure out how to complete the other part of the project my collaborators used to be responsible for. I take an online course to learn new skills and also ask for advice and information from other employees and the internet for the struggling part. After many nights of additional work, and by using the limited information, I was able to finish the rest of the project with high quality and high standard on time.
*3. Tell me about a time when you had to take a calculated risk at work (Bias for Action)*
I was working on a research project and is focused on training deep learning models to recognize user sentiments with limited examples. By chance, I know that some teams in big companies were working on a similar project. I knew that I might be behind them as I do not have enough training resources as big companies, and if they released the work earlier, I would lose many credits and recognitions for my work. So I made a calculated risk, that I decided to mainly focus on the few-shot part, which requires a few examples to train models, and I am also very familiar with this part. Instead, I skip the less familiar zero-shot experimental part where no training examples are available for the models. I quickly trained models and collected the results and analysis into a research paper, and then I released my work. My work turns out to receive lots of attention, and there are several follow-ups and good suggestions. Of course, I received some interest for the zero-shot part results, and I adopted feedbacks and added up the results soon. I think my calculated risk is worth it as my work received lots of attention and many papers follow my work. Besides, the feedbacks make my work better on the following versions.
*4. Tell me how you would complete a project or task if you had limited time or resources? (Frugality)*
I was working on a project with limited resources and time, in which the task required to traiyze their code and fix the problem. I also wrote a document with a better guideline to use the tools. Finally, there are no more issues with their data augmentation tools. I received official thanks from that team, and my manager also praised me.
*12. The biggest mistake you made and what did you learn from it? (earn trust, customer obsession)*
The biggest mistake I made is that when I was working on an intern project. I need to create a demo with live interactions with users where my model could correctly detect the sentiments of user queries. The deadline is very tight, and I need to finish it within one week. At that time, I was building a high-performance classification model, which was an essential part of the demo. I paid too much attention to the details because I wanted to make the system perfect to detect user queries. And I did not notice that I didn’t have enough time to finish all the details. The last day before the deadline, my manager pointed this out. I apologized to him, and I started to figure out a way to keep the demo as we wanted but sacrificing the model inference time that the customer won’t notice because this is just a demo. After presenting the demo, I rewrite the model inference code to meet all the requirements. Finally, the demo has a good influence on users. But from this mistake, I learned that although details are definitely important, I also need to pay attention to the whole project schedule. Besides, I also need to keep good communication and updates with my teammates and managers when I have plans.
*13. Your strength*
I am very good at planning and scheduling, I usually make a good plan for my projects and set up milestones for every week.
*14. Your weakness*
My weakness is that sometimes I would pay more attention to the details of a project and spend too much time analyzing detailed results or findings. One example is designing a multi-modal fusion model to encode different information, including texts and images. I need to get the initial results on the target dataset. While during the experiments, I spent much time analyzing whether different pre-training computer vision models affect the retrieved quality of images. Actually, we can use any pre-training models to get the feature. It takes me extra hours to finish the project. Later, I try to check myself with regular milestones and refocus on the big picture, like first designing a whole model to handle all of them instead of only focusing on small parts. Finally, I finished my work with high quality and submitted it to a top-tier conference before the deadline.
|