This is a collection of questions and answers. The questions are from Internet, interviews, or my own thoughts. The answers are mostly from my own perspective and understanding. It is inspired by my supervisor Dinh, who has a excellent capability of asking questions and answering them. I think self-reflection is the key to unlock this capability that leads me to this collection.

Questions

Why do you want to do research?

What do you want to do next? after PhD?

If you have to organize a reading group about a new topic (e.g., Multimodal learning) what will you do? (asked in an interview)

What is your most catastrophic/costly failure and what did you learn from it? (asked in an interview)

Tell me about a time when you had a confliction with your supervisor on a project and how did you resolve it? (asked in an interview)

How to engage with clients, supervisors, and colleagues?

Tell me about a time when you had an obstacle in your research and how did you overcome it?

Why do you want to do research in our group? Why do you want to do this project?

Are you a researcher or an engineer?

What is the difference between a good researcher and a great researcher?

Behavioral questions

How do you respond when you disagree with a coworker? Tell me about a time you dealt with conflict on a team. How did you solve it?

Situation: While working on a computer vision project for ID photo verification back then when I was a research engineer at Trusting Social, I encountered a technical disagreement with my supervisor regarding the approach to detect glare in photos. My supervisor advocated for traditional image processing methods, while I believed a deep learning approach would be more effective.

Task: I needed to develop a reliable module for glare detection in ID photos while maintaining a good working relationship with my supervisor and ensuring the best technical solution was implemented.

Action:

  • Implemented both approaches in parallel to compare their performance objectively
  • Developed a simple CNN model as a proof of concept
  • Collected comprehensive test data to evaluate both methods
  • Prepared a detailed presentation comparing the results, including accuracy metrics and example cases
  • Professionally presented my findings to my supervisor, acknowledging the merits of both approaches

Result: The CNN model significantly outperformed the traditional thresholding method, achieving higher accuracy and robustness. My supervisor appreciated the data-driven approach to resolving our disagreement and approved the deep learning solution. This experience strengthened our working relationship and established a precedent for resolving technical disputes through empirical evidence.

Tell me about a time you failed at work. How did you deal with it?

Situation:

I was working on a collaborative project with Australia’s Department of Defence’s DSTG and had completed a research paper for submission to WACV, a top computer vision conference. Due to the sensitive nature of the work, the paper required DSTG approval before it could be submitted.

Task:

My job was to ensure the paper received approval and was submitted before the conference deadline, while strictly adhering to DSTG’s security protocols.

Action:

  • I followed the standard procedure and submitted the paper for approval, expecting the usual two-week review timeline.
  • When delays began to occur, I regularly followed up to check on the approval status.
  • After realizing we couldn’t meet the first submission deadline, I decided to wait for the second submission round and prepared the paper for submission as soon as approval came through.

Result:

  • The paper was submitted in the second round and received generally positive reviews with minor, easily addressable feedback.
  • Unfortunately, the second round did not include a rebuttal phase, and the paper was ultimately rejected.
  • This experience taught me several valuable lessons, such as starting the approval process much earlier for sensitive projects, being more proactive in seeking guidance from senior collaborators, exploring alternative submission options like abstract submissions, and better planning around institutional requirements for future submissions.

Though the paper wasn’t accepted, I used these insights to improve my approach for future collaborations, ensuring smoother submission processes in later projects.

Tell me about an interesting project you’ve worked on recently.

Situation:

I recently worked on a project to develop an erasure method for removing unwanted concepts, like nudity or guns, from foundation models, particularly Text-to-Image Diffusion Models. These models are being widely used in real-world applications, and the risk of misuse is a real concern. This project felt meaningful to me because of its potential to address a serious societal issue.

Task:

The goal was to create a simple yet effective method to remove these unwanted concepts while making sure the approach could be practically applied. Along the way, I wanted to contribute to the research community by sharing my learnings and helping others working on similar problems.

Action:

To tackle this, I:

  • Dived into research: I did a thorough literature review to get up to speed on existing work. To make the process more efficient (and useful for others), I wrote a blog post summarizing my findings.
  • Documented challenges: While testing and implementing other methods, I kept track of the challenges I faced and shared those online. It helped me stay organized and provided useful insights for others in the field.
  • Built and refined a method: After understanding the gaps, I developed a new approach that turned out to be both simple and very effective.

Result:

In just a year, I managed to publish three papers on this topic, and one of them was accepted at NeurIPS 2024. I’m proud of this project because it not only solved a practical problem but also opened up interesting insights into how erasing one concept impacts others. Plus, it showed I could handle a fast-paced and impactful project successfully!

Tell me about a time you faced a really hard problem / a challenge at work

Situation/Task: I want to develop a theoretical framework to understand why and how personalization methods can learn from reference images. My hypothesis is that each image has a combination of personal features (such as your specific hairstyle, face shape, eye color, etc.) and generic features (such as the background, context around you, etc.). The current approach is just simply learning the mean/center of the personal features in the latent space. While it looks simple, but it comes with many implications and intuitions, for example,

  • If there are two people in the training set, the learned code is the average of the two, or might represent both of them.
  • If there is not variance in the personal features, the learned code is also less variance, meaning that you cannot generate different styles/views of that person.

Problem: I don’t have a solid background in mathematics. I have intuition about the problem, the final result, but I don’t know the path to get there.

Action: I believe in the principle that “the luck comes to prepared mind”. I indicated the gap in the problem and waited for the opportunity to fill it. It helped me to navigate when reading papers, and anchor my thoughts around the problem whenever I read a new paper. Such as, “Does this paper fill the gap?”

Result: Until now, I still don’t find the exact solution to the problem, but I have a pretty good pathway to get there, inspired by another paper from CMU.

Tell me about a time you needed information from someone who wasn’t responsive. How did you handle it?

Situation:

Task:

Action:

Result:

How do you deal with unexpected changes to deadlines? Tell me about a time you had to meet a tight deadline.

Situation:

Task:

Action:

Result:

Tell me about your biggest weakness.

Tell me about a time you adapted to a new situation or environment.

Situation:

Task:

Action:

Result:

Tell me about a time you had multiple responsibilities to manage. How did you respond to this situation?

Situation:

Task:

Action:

Result:

Why do you want to work here?

Common themes

  • Conflict resolution Example: How to resolve the technical disagreement with my supervisor regarding the approach to detect glare in photos.

  • Tight deadlines: Try to manage the time effectively (calendar, split to smaller chunks, etc.), prioritize the most important tasks. Ask for help. Example: How to collaborate with Trang and Long to submit a paper to NeurIPS 2024.

  • Mistake/failure Example: Fail to seek approval from DSTG, leading to missed the paper submission deadline.

  • Leadership or influence Influence through writing, willing to share the knowledge and help others. Example: How to write websites and blogs

  • Learning something new quickly Learning through writing. Not only theoretical knowledge, but also how to implement it. Example: How to learn the theory of Recursive Reasoning and apply it to the Relation Extraction problem.

  • Working under ambiguity Example: How to find the idea for the Personalization project while most of the time, was working on the Machine Unlearning project.

Amazon Leadership Principles

Customer Obsession: Leaders start with the customer and work backwards. They work vigorously to earn and keep customer trust. Although leaders pay attention to competitors, they obsess over customers.

  • Example:
    • As a Teaching Assistant, I always try to understand if I was a student, what I would expect from the lab/unit. Then, I would try to prepare the lab materials to convey these knowledge in the best way. For example, when teaching the Deep Learning unit at Monash University, I prepared the skeleton/learning plan for all the labs, and summarized the key knowledge in each lab, and reminded the students when starting the lab. I aso design a quiz test (with 10-15 questions) for each lab to make sure the students understand the material. This quiz has been used since 2020, and it has been shown very helpful with alot good feedbacks from the students. At the end, I have received one of the best - SETU - Student Experience Teaching Feedback (very good grade).
    • As a Project Co-lead - collaboration with DoD:
      • Situation: I am a Project Co-lead, responisble for doing the research and report to DoD at our biweekly meeting.
      • Task: They usually ask similar questions, even I have presented and answered several time before.
      • Action: I try to prepare a report and send to them before the meeting. I am also write some key points that I want to highlight at each meeting.
      • Results: The DoD team has been being more engaged with our meeting, they know the progress and provide more valuable feedbacks.

Ownership: Leaders are owners. They think long term and don’t sacrifice long-term value for short-term results. They act on behalf of the entire company, beyond just their own team. They never say “that’s not my job”.

  • Example:
    • As a Research Fellow, a lab member, and a project co-leader, my main task of the project is conducting the research but not recruiting students. However, I still try my best to promote the lab to attract more students. I promote the lab in Vietnam Ph.D. forum, and from that, got several students joined/interviewed in the lab.
    • As a Teaching Assistant, even with the limited time, I always try to find the best way to convey the knowledge to the students. I take the responsibility that I am the one who will teach them and make sure they understand the knowledge. Even, sometimes, the lecturer advised me just to given the students a quick overview of the lab and let them play with the code, I still try my best to convey the basic background related to the lab so that they can understand not only the code, but also the basic knowledge behind the code.
    • As a project co-leader, I have a responsibility to make sure the project is on the right track.

Invent and Simplify: Leaders expect and require innovation from their team and always find ways to simplify. They are externally aware, look for new ideas from everywhere, and are not limited by “not invented here”. As we do new things, we accept that we may be misunderstood for long periods of time.

  • Example:
    • As a project co-leader, I have a responsibility to make sure the project is on the right track. It includes finding the potential ideas, sometimes, even the new/potential problems. For example, I found that the loophole of the Textual-Inversion method that makes it not as good as the Dreambooth method, i.e., because of the tokenization and the way to inject the prompt/control signal into the generation process.
    • Teaching Assistant - Simplify the material - so that students can understand not only the code but the background.
      • Situation: I am a Teaching Assistant, responsible for teaching the lab materials but not for the background of the lab.
      • Task: To help the students understand the background of the lab.
      • Action: Have a list of items that we have learned each week, convey the main points/contents that will be covered in this lab. Design a quize test to check their understanding.
      • Results: Got high SETU feedback

Are Right, A Lot: Leaders are right a lot. They have strong judgment and good instincts. They seek diverse perspectives and work to disconfirm their beliefs.

  • This principle is a bit confusing
  • Example:
    • Seek diverse perspectives and work to disconfirm their beliefs.

Learn and Be Curious: Leaders are never done learning and always seek to improve themselves. They are curious about new possibilities and ideas.

  • Example:
    • List of ideas that I would do if I have more resources or time. Try to learn and fill the gap in my knowledge toward the problem.
    • Idea about Real-Estate Agent:
      • Crawling the data from the internet, realestate.com.au, domain.com.au, etc.
      • However, these websites have mechanisms to block the scraping.
      • Recently, I found that we can use https://github.com/ScrapeGraphAI/Scrapegraph-ai to automatically generate adaptive scrapers for different websites, playing as a normal user to bypass the mechanism.

Think Big: Thinking small is a self-fulfilling prophecy. Leaders create and communicate a bold direction that inspires results. They think differently and look around corners for ways to serve customers.

  • Example:
    • Aim for becoming a top researcher in the field of Machine Unlearning.
      • Collaborate with other Professors from Monash University, to build a tutorial/workshop about Machine Unlearning.
    • Maintain a list of potential ideas/problems that I want to solve.

Bias for Action: Speed matters in business. Many decisions and actions are reversible and do not need extensive study. We value calculated risk-taking.