The Cornell, Maryland, Max Planck Pre-doctoral Research School 2023 Observations and Gained Insights

Group Photo

Panel Session 2: "Research in industry vs. academia"

Problem Focus & Recognition

• Industry tends to focus on concrete problems.
• In academia, broader issues are often addressed.
• Authorship and credit in academia is complex. It's not a zero-sum game. It's not just about who is first or second author; giving credit to students doesn't mean professors won't get any.

Publication & Quality

• The emphasis is on publishing fewer papers but ensuring they are of high quality. It's not about the quantity but the impact and quality of the papers.
• The first, last, or best paper on a topic are the most influential.

Career Path

• Before securing a tenure professor position, many go through multiple postdocs and even stints as industrial research scientists.
• Only about 10-20% of PhDs eventually become faculty.
• Some research scientists find academic-like environments within the right industry groups.

Factors Differentiating Academia and Industry:

• Industrial research must eventually have some commercial value.
• In academia, there are constraints like obtaining funding, student recruitment, and equipment acquisition.
• Academics have better job security and can rebound from mistakes.
• Industry doesn't need to chase grants or funding in the same way academia does.

Skills & Transitions

• Transferring skills between departments or companies is straightforward.
• Transitioning between academia and industry is often a one-way street. It's challenging to return to academia from industry unless one maintains a consistent publishing record and works on research-valued projects.

Geographical and Topic Mobility

Researchers are encouraged to be flexible, moving across countries and topics.

Work-Life Balance

• Systems vary across locations.
• Enforce personal boundaries and learn to say no.
• A balance doesn't mean absence of stress. In the industry, even if the work-life balance is okay, stress may arise from working on undesired projects or facing peer pressure.
• Find people who become friends with you.

Two-Body Problem

It's more of an issue in academia than in industry since it's easier to change companies than academic institutions.

Personality and Approach

• Industry caters to hackers and those interested in tooling.
• Academics focus on research and higher purposes and see coding as a tool. Effective communication, including selling your idea in proposals and talks, is vital.

Startups vs. PhD Journey

• Both require a significant commitment, typically around 6-8 years to IPO.
• Startups demand full devotion, often with little to no work-life balance.

Funding & Tenure

• If a grant from a company fails, there will be no direct legal consequences, but the likelihood of getting another might be reduced.
• Tenure provides a basic salary and job security, but researchers still need to raise funds for their research.
• Doing a job aligned research can be beneficial for dissertation and future career opportunities.

Laxman Dhulipala (2nd Lecture)

• Graphs are ubiquitous structures. Implementing high-performance graph algorithms speeds up scientific discovery.
• I don't work on dense graphs. Real-world graphs are sparse, and I haven't seen a dense graph in practice in 10 years.
• I focus on shared-memory algorithms and don't recommend programming supercomputers until you have to.
• Recommended reading: Scalability! But at what COST?
• Should batch updates to dynamic graphs
  • More parallelism
  • Reduces the cost of each update
• Representing adjacency information using purely functional trees are safe for concurrency.

Guest Lecture: Yiting Xia

• There are different available connections at different time slices.
• Precomputing routes and handling link failure is still work in progress.

Group-Mentoring Session

Peter Druschel and Bobby Bhattacharjee

Key Skills and Knowledge

• Emphasized the importance of academic aptitude and the ability to work in unstructured environments.

Problem-solving Approach

• Seek problems that are significant, solvable, and align with your skill set.
• Recognize that one may not always approach the right problem from the best angle.
• Handling setbacks is crucial. Time spent on tackling a problem is never lost.
• Resilience, dedication, and discipline are essential traits for success.
• Read many things that are loosely related to solve a problem, as they might offer insights.

Application Strategy

• Apply to a minimum of 5-10 institutions.
• Do the homework for providing a strong application, especially given low acceptance rates, like 10%.

Interests and Graduate Programs

• Have a broad range of interests when considering a graduate program.
• Opt for programs that offer a wide variety of choices.
• Expressing diverse interests in applications can improve acceptance chances.
• It's advisable not to close one's doors apriori.

Monitoring Progress in Grad Programs

• A competent group advisor is crucial, as they will guide and look out for students challenges like selecting an excessively challenging problem, lacking motivation, or poor time management.
• Set achievable milestones that lead to publications, helping to build a solid publication record. ### Mariya Toneva

Changing Discipline during Ph.D.

• Evaluate if the institution has the necessary resources to support this transition.

Traits of an Ideal Ph.D. Student

• Effective communication skills.
• Strong critical thinking abilities.
• A robust computational background.
• Prior research experience.

Linguistics

• Noted a resurgence in the domain of linguistics as opposed to pure data-driven techniques.

MPI-SWS

• MPI-SWS is highly recommended for programming languages, especially when collaborating with diverse groups of people.

Diving into NLP (Natural Language Processing) - Hop On Now?

• When considering venturing into NLP, focus on:
  • Experts who have a distinct vision in a less-saturated niche.
  • Those with substantial experience in related fields, such as the intersection of NLP and robotics.

Distinguishing Yourself in Applications

• To stand out:
  • Foster qualities like initiative, drive, and ambition.
  • Accumulate experiences that align with and support your academic and research interests.
  • Obtain references that can vouch for your character and work ethic.
• It's also essential to explore and consider multiple options or paths.

Lorenzo Alvisi

Cultivating an Academic Sense

• To nurture an academic mindset, one should assess how an individual performs when faced with a problem.
• He mentioned the "Dijkstra club" at UT Austin as an example.
• Emphasized the significance of "beautiful work" and that it's crucial for individuals to produce work of beauty and quality.
• Observing and learning from the endeavors of others is beneficial.

Life's Blueprint

• Life does not come with a set map but rather a compass for direction.
• Professor Alvisi never limited his imagination about his capabilities.
• Guiding principles in life:
  • Seeking personal happiness.
  • Maintaining healthy relationships.
  • Pursuing a fulfilling job that combines happiness with challenges.
• Acceptance of uncertain outcomes: One might not always know if they will succeed or fail.
• The importance of personal growth: Find joy in self-improvement.
• Shared personal experience of pursuing two Ph.D. degrees, the first of which was at an institution he didn't particularly favor. Highlighted that struggles are often hidden from view.

Career Perspectives

• One's career doesn't necessarily peak at a fixed point; there's always potential for growth, including entering academia.
• Career choices are not always black and white; it depends on personal preferences and aspirations, such as seeking excellent opportunities close to home.
• Consider the duration of your investments in particular career choices. Not every commitment needs to be long-term.

Balancing Hobbies and Work

• Prof. Alvisi shared advice from his mentor's mentor about integrating hobbies into professional life.
• While he had diverse interests, he made sacrifices to focus on computer science due to his intellectual capacities. Some hobbies were too time-consuming.
• Emphasized the importance of hobbies as they provide a necessary balance and maintain mental well-being.

Addressing the Two-Body Problem

• Universities recognize the challenge when both partners in a relationship are professionals.
• If partners excel in different domains, there's potential for both to be hired with attractive incentives.
• Solutions include proactive planning, alternating priorities between partners over the years, and considering remote work opportunities.

Other Insights

• Mentioned the Sloan Fellowship as a notable achievement before tenure.
• Advised young professionals to delay specialization as long as possible. Explore various options.
• Encouraged students to seek advice from multiple professors to gain a diverse range of opinions and insights.

Tapomayukh Bhattacharjee (2nd Lecture)

• There are six activities of daily living (ADLs) defined in literature: personal hygiene or grooming, dressing, toileting, transferring or ambulating, and eating
• Anomaly detection is used in processing sensor data.
• A* is widely used in motion planning due to its efficiency and optimality (it never overestimates the cost).
• Motion planning time = search time + collision checking time (~90%). Therefore, the author proposed lazy A* (which finds an optimal path in an unconstrained situation, goes over collision checking while on the path, and re-searches a path if a collision is encountered).
• Collect a dataset before embarking on research.
• To understand how to manipulate different kinds of foods, the author created a food manipulation taxonomy.
• Choose hardware components for real-world deployability.
• Use deformation of points on a gel coupled with computer vision algorithms to measure shear force
• Add structure to machine learning algorithms to overcome a lack of data.
• If integrating multimodal data sources, think of where to integrate as the size or magnitude of different data may be inconsistent.
• A "bandit" algorithm is an RL algorithm where we utilize partial feedback of one step in the decision-making process, unlike conventional RL algorithms with "episodes" spanning multiple steps.

Audience question: How to stay up-to-date with the state-of-the-art (especially in the fast-changing landscape of machine learning)?

• One of the main tasks of faculty life
• Look at titles and abstracts of publications in all well-known conferences.
• Organize reading groups and reading sessions.
• Interact with known other research groups.

Derek Dreyer: How to write papers and give talks that people can follow

• Many papers suffer from the TMI (too much information) problem.

• Aim at giving constructive principles that are easy to check and fix.

• A paper is different from a textbook - people aren't as committed to reading a paper as they are to reading a textbook.

• A good but not interesting paper tends to get a "B" or a "weak accept."

• Putting the Related Work section at the front (as opposed to in the back before the Conclusion) may hinder unfamiliar authors from understanding your work.

• Most people don't listen to talks to determine whether they should read a paper. Instead, they listen to talks to discuss with others. The main goal of a talk is to give people positive feelings about your work.

• A talk should only cover the intro and key ideas sections of the corresponding paper.

• The key ideas should be the high point in your talk before presenting the takeaway messages.

• Add visual elements to emphasize one point per slide.

• Use smooth animations to help the listener follow.