What is a PhD Programme Like? My PhD Student Experience Part 2

TL;DR

The marathon-time prediction project relied on converting 16 weeks of wearable-sensor training data (including 100-meter intervals) into model-ready features such as weekly distance and session counts.

Briefing Cornell Notes

Briefing

The first year of one PhD program turned into a fast-moving mix of coursework, early research output, and major life disruption—then reshaped again by a shift to working from home and a new research angle. After a mid-January push to get a paper submitted quickly, the work focused on predicting marathon finish times from the 16 weeks of wearable-sensor training data (including 100-meter interval information). The project required building features from raw training logs, selecting signals informed by sports-science literature, and implementing a case-based reasoning modeling approach. With only about four weeks for programming and data work and another week or two for writing, the effort culminated in a submission in early March, followed by the familiar late-night cycle of graph iterations and last-minute adjustments before acceptance.

That research sprint mattered because it built both technical confidence and a clearer writing workflow. The supervisor provided close support on where to start with the write-up and how to present results, while leaving much of the data conversion, feature engineering, and model implementation to the PhD student. Even after submission, the process left a mark: long weekends, slow-running computations, and repeated figure revisions—especially because the supervisor had a stronger sense of how graphs should look. Once the paper was submitted, a planned trip to Italy was canceled as the pandemic intensified in Dublin; a short break in Belfast became a last pre-lockdown memory before full lockdown closed everything for months.

The pandemic then forced a practical reorganization of academic life. Classes and teaching responsibilities moved online, and the first attempt at teaching hours online proved difficult before becoming more manageable later. Working from home, the student set up a dedicated workspace in a renovated part of the house—an advantage made possible by finishing a kitchen remodel just before lockdown. With classes completed and the home routine stabilized, research pivoted toward recommendation-focused work tied to marathon prediction. A short paper for the conference RECSYS (targeting “recommender systems”) used a more explainable modeling strategy by generating recommendations from single-week slices of training data, enabling clearer links between recommendations and individual runners. Submission landed by late May.

Alongside the academic workload, the year included entrepreneurial and professional development. In summer, the student and a fellow PhD student friend, Courtney, entered the Nova UCD Student Enterprise Competition with a productivity app concept aimed at personalized productivity for people working from home. A four-week entrepreneurship course guided customer discovery through polls and DMs, prototype testing via workshops, and a final pitch to investors; the team placed second and won €1,000 each. The same summer also brought a heavy personal turning point: the death of the drama school director the student had worked with for years, leading to the student becoming the only director and taking on full management duties during evenings and weekends.

By the end of the first year, the student had also presented at the International Conference on Case-Based Reasoning (including a 20-minute talk with Q&A) and submitted a doctoral consortium paper for feedback from specialists. Research output continued through a journal paper where the student contributed a case study, and a broader collaboration began with universities in the UK and New Zealand (University of Hertfordshire and Auckland University of Technology) to validate lab-measured fitness variables against predictions from marathon training data. YouTube monetization also started, adding a modest but meaningful income stream—while reinforcing that the channel’s growth required sustained effort rather than quick returns. The throughline: early research momentum, rapid adaptation to pandemic constraints, and a widening set of responsibilities that demanded constant prioritization.

Cornell Notes

The PhD student’s first year combined an aggressive research push with major pandemic-driven changes and expanding responsibilities. Mid-January work began on predicting marathon finish times from 16 weeks of wearable-sensor training data, using feature extraction and a case-based reasoning approach; tight deadlines required rapid programming, then iterative writing and figure refinement. After lockdown forced classes and teaching hours online, research shifted toward a more explainable recommendation-system paper for RECSYS, generating recommendations from single-week training slices. Summer added entrepreneurship (Nova UCD Student Enterprise Competition) and heavy personal workload when the drama school director died, leaving the student as the only director. By year’s end, the student had conference presentations, a journal co-authorship, and an international collaboration to validate fitness predictions with lab measurements.

How did the marathon prediction project turn raw wearable data into something a model could learn from?

The work started from 16 weeks of training data collected via wearable sensors (including Fitbit-style inputs) and raw 100-meter interval information. The student had to convert that raw time-series into machine-learning features, guided by both the supervisor and a sports-science co-supervisor. Feature ideas included total weekly distance and the number of sessions per week, with additional feature development informed by prior sports-science literature. Modeling used a case-based reasoning approach from a prior paper, but the student implemented the pipeline and learned how to make the data usable for machine learning.

Why did the later RECSYS paper emphasize explainability, and what changed in the modeling?

The follow-up work targeted recommender systems (RECSYS) by changing the recommendation mechanism so it could point to specific training evidence. Instead of using broader context, the model generated recommendations using single weeks at a time. That design made it easier to connect recommendations to individual runners’ training segments, improving interpretability compared with the earlier marathon-time prediction setup.

What did the pandemic disrupt most during the first year, and how did the student adapt?

Lockdown shifted coursework and teaching hours online, and the first online teaching experience was described as genuinely difficult. After about a month of settling into remote work, the student adjusted routines and completed class obligations while continuing research. Working from home became workable because a renovated area of the house (including a kitchen completed just before lockdown) provided a dedicated workspace, even though foot traffic from shared household areas still created interruptions.

What non-academic commitments shaped the student’s time in summer, and what impact did they have?

Summer included two major demands. First, the student and Courtney pursued the Nova UCD Student Enterprise Competition, spending weeks on customer discovery, prototype development, and pitching; they placed second and won €1,000 each. Second, the death of the drama school director led to the student becoming the only director, effectively turning evenings and weekends into management work. That reduced available time for PhD research and also required clearing and storing decades of school materials.

How did the student build research credibility early—beyond journal submissions?

The student presented at the International Conference of Case-Based Reasoning with a 20-minute talk plus 10 minutes of Q&A, marking a first major conference presentation as a PhD student. The student also submitted a doctoral consortium paper for a panel review, aiming to get feedback from researchers outside the immediate niche of the supervisors’ expertise. The student valued the consortium as a way to avoid getting “boxed in” and to pressure-test plans with domain-adjacent specialists.

What international collaboration began during the first year, and what was its purpose?

A collaboration started with University of Hertfordshire (UK) and Auckland University of Technology (New Zealand). The goal was to estimate fitness variables from marathon training data, then validate those predictions against lab-based measurements from sports-science trials. Because lab data collection happened with one runner every couple of months—and pandemic delays occurred—the work continued across the year and into the next.

Review Questions

What specific steps were required to transform raw 100-meter interval training data into features suitable for machine learning in the marathon prediction project?
How did the modeling strategy change between the marathon prediction work and the RECSYS recommendation paper, and how did that affect explainability?
Which summer activities competed with PhD research time, and what concrete outcomes came from each (e.g., competition placement, conference participation, collaboration milestones)?

Key Points

1
The marathon-time prediction project relied on converting 16 weeks of wearable-sensor training data (including 100-meter intervals) into model-ready features such as weekly distance and session counts.
2
A case-based reasoning approach was implemented from prior work, but the student handled most of the data conversion, feature engineering, and modeling execution independently.
3
Pandemic lockdown forced teaching hours and classes online, with early online teaching described as especially challenging before improving after a settling period.
4
Research direction shifted toward recommender systems for RECSYS, using a single-week recommendation strategy to make results more explainable and tied to specific runners.
5
Entrepreneurship during the summer included the Nova UCD Student Enterprise Competition, where the team placed second and won €1,000 each after building a personalized productivity web-app prototype.
6
A personal leadership transition at a drama school created a sustained management workload, limiting time for PhD research and YouTube posting during parts of the year.
7
By year’s end, the student had conference presentations, a doctoral consortium submission, a co-authored journal paper contribution, and an international lab-validation collaboration with University of Hertfordshire and Auckland University of Technology.

Highlights

The marathon prediction pipeline required turning raw 16-week wearable training logs into features and then implementing a case-based reasoning model under a tight submission timeline.

The RECSYS paper’s explainability came from generating recommendations from single-week training slices, making it easier to link recommendations to individual runners.

Lockdown didn’t just change logistics—it reshaped the research schedule, teaching responsibilities, and even how the student organized workspace at home.

The student balanced PhD research with entrepreneurship and sudden drama-school leadership, illustrating how academic progress can depend on constant triage and adaptation.

Topics

PhD Year One
Marathon Prediction
RECSYS Recommendations
Teaching Online
Case-Based Reasoning
Entrepreneurship Competition
Pandemic Remote Work
Lab Validation Collaboration