Goodbye, tenure track

I wasn’t sure about sharing this, but in the original spirit of my blog, that I ought to. 

I am leaving my tenure track position. 

There it is. It feels good to write it down. There are a lot of failure related thoughts here, which I will be sharing in future posts. But first, a bit of background about what happened. 

Tenure criteria 

In the previous post I wrote about starting my tenure track position and what I was planning to achieve in 4 years. To recap, here is summary of the goals, which were approved by the department

  • Get teaching certificate
  • Setup and teach a course, co-teach in other courses
  • Supervise at least 2 MSc and 4 BSc students
  • Co-supervise a PhD researcher
  • Co-author of at least 5 peer-reviewed publications in high impact, relevant journals
  • Setup collaborations with other departments 
  • Apply for 2 medium-sized (1 PhD or postdoc) grants per year
  • Apply to small grants, for example for workshops, when possible
  • Give talks at (local) conferences, or invited talks if possible
  • Outreach about academia through blog and Twitter 

Progress so far

As far as teaching goes, all goals are achieved. I setup a course, taught in another course (both 3 years in a row now), and recently gave a number of lectures in a MSc course. So far I supervised 5 MSc students and 12 BSc students. I’m the daily supervisor of two PhD researchers, one based on my own funding efforts.  I also received my university teaching qualification in 2019. 

Research-wise, things are alright. I published six journal papers and one preprint, but it could be argued that some of these do not count. For example three were started during my postdoc, although I put in more hours during my tenure track. There’s also the Twitter paper, which is not on the topic of my research, but probably has had more impact than the others combined. I am also quite happy with my Google scholar numbers.

I am not sure about the funding. I applied for two larger grants per year as agreed, and 1 of these was funded. The others are in my failure CV. This is in line with the overall success rate, and several smaller grants were funded as well. But I have the feeling this is not sufficient, even though the tenure criteria do not specify it.

In terms of visibility, things are good. Especially in the first two years when I was blogging regularly, my website and Twitter were growing steadily. I think this has contributed to invitations for talks, and I have given more talks, including international ones, than I ever expected. I’ve also been invited as an associate editor, social media chair and other similar roles. 

So overall, not bad, considering that in my third year I was seriously ill and I spent several months recovering, which was extremely difficult. Even so I did get a few things done in that time, such as the teaching certificate. Overall, things could have been better, but given that I had no start-up nor PhD researchers I could co-supervise from the start, overall I’m actually quite happy with what I achieved.

Perhaps here I should mention two other developments. The first is the artificial intelligence “brain drain” in the Netherlands, limiting the number of people willing to teach. The second is a position paper by several organizations (including funding agencies), that aims to redesign how researchers are evaluated, and to recognize factors other than the h-index. Music to my ears.

Midway evaluation 

As I explained in the previous post, traditionally there is a midway evaluation halfway through the tenure track, to see what else is needed to fulfill the tenure criteria. My midway evaluation was scheduled for May 2019, but a month before that I became ill, so this was cancelled.

Towards the end of 2019 I was working full-time again. The idea was to schedule an unofficial midway evaluation, a year ahead of the final evaluation. I gave a talk about my research and updated my CV and progress document (summarized above).   

Given this information, the committee advised that, I will probably not get tenure if I have the final evaluation as planned in 2021. The proposed solution was to give me a temporary contract and have the final evaluation later, so that I have more time for, between the lines, getting funding and writing more papers. 

Tenure clock extension, that’s good right?

Although to many readers this extension might sound good, I declined the offer. I will therefore be leaving my tenure track position.

The first reason for this decision is the uncertainty. I believe that the trigger for my manic episode was staying up at night to write grants, and I don’t want my life to depend on a lottery. There is also no definition of what “enough” would be, and that once I achieve those things, I would get tenure.

Secondly, I feel like my illness is a bad excuse that there wasn’t enough time to evaluate me. But people are at times evaluated after two or three years – researchers who are employed by the same university before starting a tenure track position, due to the labor laws.

But most importantly, I don’t want to be in a place with such priorities. I have achieved most goals on my list – goals that were agreed upon at the beginning – despite having a major illness. I will not be an award-winning researcher, but I feel – and people have told me – that the things I do are valuable. If the university does not see this, I need to find a place that does.

What next?

My current contract runs out at the start of 2022, but since I made this decision already, I will probably leave earlier.

For now I will be finishing up various projects, and slowly searching for a job.

So dear readers, I am now officially open for job opportunities! I don’t want to limit myself to specific job titles or sectors just yet. So if you think you could use my research, teaching, outreach, organizing, blogging skills (academic CV here), please get in touch.

That’s it for now, but expect more failure-inspired content soon!

Tenure track in the Netherlands

By popular demand, today’s post is about my tenure track position which I started 3 years ago. Although I intended to give an update of how my tenure track is going, there’s a bit of background that’s relevant to share, so this post is only about my experiences when I started. Also, recently I’ve had a few questions from future tenure trackers, so I’m sharing my answers in case it is useful to others.

Starting conditions

As I’ve also explained in my “student or employee during your PhD” post, all academic positions in the Netherlands work with fixed pay scales. You can find these here, below I also added a screenshot of some of the scales.

These numbers are all before tax and per month. Various secondary benefits also apply.

Assistant professor positions are in scale 11 or 12. Typically a starting assistant professor would be in scale 11, and in scale 12 after tenure. The Dutch Network of Women Professors (LNVH) reports that 50.8% of women assistant professors are in scale 11, versus 40.8% of men.  

When I started at TUe, I was initially offered scale 11.0. However, I had already been in scale 11 as a postdoc, and my institution was a medical center, with slightly higher pay scales. Due to this I was offered 11.3, which just matched my previous salary, and which I accepted.   

There was no start-up package – I think this in general isn’t a thing in the Netherlands, although I do see this being offered more frequently now.

Contract & tenure conditions

The tenure track contract is a temporary contract for 5 years. After 4 years there is an evaluation which decides whether you get tenure or not. If yes, you get permanent contract, if not, you are still employed for a year. There is also a (less formal) midway evaluation after about 2 years, to prepare for the real thing.

The criteria for evaluation are described in various documents. I received some general criteria on what is important for the university (for example “supervising students”), and a department-specific interpretation of these criteria. In the context of creating a personal development plan for the tenure track period, I did receive some quantifiable criteria too, of what you should aim for within 4 years:

  • Significant progress in obtaining the teaching qualification certificate
  • Responsible instructor for 1-2 courses
  • Good teaching evaluations
  • Supervision of at least 2 MSc and 4 BSc students
  • Co-author of at least 5 peer-reviewed publications in high impact, relevant journals
  • Written statement from chair about contribution to getting funding
  • Significant progress in increasing external visibility
  • Collaborations with other departments, hospitals or industry
  • Successful (co-) supervision of multiple PhD researchers
  • Examples of strong leadership
  • Examples of strong communication skills
  • Examples of independence and responsibility

A bit more quantifiable, but still open to interpretation. In my own personal development plan I translated these as follows:

  • Get teaching certificate
  • Setup and teach a first year course, co-teach in a third year course, later start developing course closer to my research
  • Supervise at least 2 MSc and 4 BSc students
  • Co-author of at least 5 peer-reviewed publications in high impact, relevant journals
  • Apply for 2 medium-sized (1 PhD or postdoc) grants per year
  • Apply to small grants, for example for workshops, when possible
  • Give talks at (local) conferences, or invited talks if possible
  • Setup collaborations with other departments 
  • Co-supervise a PhD researcher (if funding)
  • Outreach about academia through blog and Twitter 

Also not entirely quantifiable, but I also left out a few specific details here (examples of papers, collaborations, numbers of blog/Twitter followers etc).

The (midway) tenure evaluation moments consist of submitting an update of this plan and a recent CV, and then giving a presentation about your achievements to a committee of 3-4 professors.  

This is all I wanted to share for the first part of this topic – next time I’ll talk about how things are going so far. If you have any questions about this post, or anything I can address next time, please comment below!

“Avengers for Better Science” has made me a Better Human

Last year together with Aidan Budd, Natalia Bielczyk, Stephan Heunis and Malvika Sharan we organized the Avengers for Better Science workshop. This guest post has been written by one of the participants, Cassandra van Gould-Praag, reflecting on this workshop.

Cass (@cassgvp) is a postdoctoral researcher at the University of Oxford Department of Psychiatry. She provides support for (f)MRI experimental design and analysis in the investigation of treatments for mood disorders. In this role, she has to stay up to speed with the leading edge of analytic tools, and is constantly on the lookout for tips, tricks, and techniques to make this research quicker, slicker, and more effective. This goes hand-in-hand with making the research more transparent and reproducible, and freely sharing the outputs of our labour. She is a contributor to The Turing Way and works with the Wellcome Centre for Integrative Neuroimaging Open Community Team. She is a passionate believer in accessibility and the equitable dissemination of knowledge, and spends a lot of time showing people that programming isn’t scary.

“Avengers for Better Science” has made me a Better Human

Avengers for Better Science” was unlike any academic event I have been to in my 10 year academic career. It will henceforth be my benchmark for collaborative, interdisciplinary, in-person, professional interactions, and a working demonstration of the level of compassion, empathy, understanding and genuine desire to “be better” which is necessary to create the type of research environment I want to be a part of.

I firmly believe that the best tool available to researchers for improving our understanding of the world is to increase the reproducibility of our research. Reproducibility goes hand-in-hand with increasing the diversity of the people who can attempt to reproduce our research; if the only people who can reproduce my work are people who are similar to me, then my work is not reproducible. The added bonus of improving the diversity of contributors is a larger potential reviewing pool. The more eyes which look, the more diverse the viewpoints which can be drawn on to solve problems, and the more likely they are to pick up errors or suggest improvements. This way of thinking underlies the “selfish reasons” to be mindful of inclusivity in research.

I try in my daily life to be aware of issues of inclusivity, but this is not for selfish reasons. This is because life is hard, and for some people life is extra hard, and I’m not about adding to the discomfort. You might call me a “Social Justice Warrior”, and I’d be fine with that. Our society deserves justice and I’m prepared to go into battle. 

The skillfully crafted program of talks and events at Avengers allowed me to demonstrate the value of understanding my own privilege as a white cisgender heterosexual non-disabled person. It also compounded the understanding that my own experience of the world may be very different to someone else’s. This position is supported by my empirical research on perception and conscious experience (for example exploring the experience of synaesthsia) which supports the idea that there is no reality except that which we perceive, and everyone’s perception is personal. 

Despite my pre-existing understanding, I had ample opportunity to learn at Avengers. I was challenged on my assumptions, reminded of the ethical imperative to be kind to myself if I want to do my best work, taught how to support others (and myself) at times of crisis, given some excellent productivity tips, and convinced for the first that there is a research environment which exists outside of academia that I could thrive in. I was also made aware of some ethical concerns in how we practice research, for example in the use of biased artificial intelligence to inform criminal sentencing, and ideas of situatedness when we consider who is leading the agenda on transparent and reproducible research. 

All of these lessons wildly exceed anything I learnt in institutional “Professional Development” courses. This was in no small part due to the excellent leadership demonstrated by the organisers as they all enacted the core values of community and inclusivity which they were aiming to foster within attendees. They worked tirelessly to build a safe space to explore our strengths and weaknesses, and made it abundantly clear that it was “OK” to be vulnerable and less than perfect. This is a lesson which is sorely missing in academia. They helped us to remember that we are all human, and that is an excellent thing. 

The funded travel and accommodation for the workshop meant I didn’t have to work too hard to justify attendance to my department. If I had, I may have struggled to define how “learning to be a better human” would help me do better research. I now understand that acknowledging my humanity makes it easier to accept my mistakes and those of others. This makes me far more open to constructive criticism, which in turn makes it a lot easier to ask questions and comfortably share my code and data. It also helps me to hold my beliefs lightly, which may reduce the bias I bring to analysis. 

An improved understanding around issues of inclusivity allows me to interact more effectively with our volunteer participants, design more ethical research and have a greater awareness of the ethical impact of our work and that of others. It also makes me a better colleague and teacher. I work harder to listen to my colleagues and students, and place more value in their truth. This makes the process of collaborative research (which all research is) much more efficient, effective and enjoyable. I’m also trying to to lead the culture change which is necessary for a healthy academia by taking care of myself, managing my own expectations and that of others, while openly and directly challenging behaviours which violate the rights of others. I am more productive now I understand my own limiting beliefs and am able to communicate my requirements with confidence.

Success in academic research is in part governed by “who you know”. I am therefore sincerely thankful to the organisers and attendees of Avengers for the community that we built at the event and networks which we continue to strengthen. Through open and inclusive research projects I know that it is possible to work as part of a team with shared values, and this usually makes for a pretty fun and productive project. I know that the connections I made through attending Avengers will stay with me throughout my career, and I am excited about the opportunities for collaboration this brings. This passion and curiosity is an excellent motivator for me. I look forward to the next opportunity to learn from my kind and diverse colleagues.

Using Evernote vs Todoist as your todo list

I’ve been a happy user of both Todoist and Evernote for a few years now – see my post on using Todoist and Evernote together with Google Calendar to get things done. 

However, last year following a period of illness I’ve reconsidered the tools I use. In this post I explain why I switched from Todoist+Evernote to only using Evernote, and why I later decided to go back to my trusted system. 

Downsides of Todoist

My main problem with Todoist is that it is too easy add tasks.

That might sound a bit weird. Of course the adding tasks functionality of Todoist widgets is great, and it is easy to capture all the little things you need to do. But since all tasks have the same “weight” (even if you give them different priority), your overall task list becomes too focused on not-always-important, little tasks. Although I was regularly organizing my list, just having all the other tasks there was kind of weighing on me.

A related problem is that when you add a task, you don’t see what tasks you already have scheduled. So you can be too optimistic when adding a task for “tomorrow”  when you already have various meetings and other tasks scheduled. 

Finally, Todoist has a desktop app, but it doesn’t work if you start it when you are offline. 

Evernote as your todo-list

Evernote is not a specific todo-list app, but it is possible to use it as such. You can see notes as individual todos, and then organize them via notebooks or tags, or you can create a checklist in a single note. I decided to go with the checklist approach, and created two notes  – “Current” and “Maybe”.  “Current” was for anything that was coming up, and “Maybe” for projects that I might or might not do. 

Most of the time I worked with the “Current” list, where I made a table with one row for each week, and columns for different types of tasks. I started with “work” and “home”, but later split these up into more categories, based on priority.

This system had several advantages that I missed with Todoist. When adding a task, I had to add it to a specific row, so I would already see what other tasks I had planned for then. Also, I became more aware of the weight of the different tasks, and I feel that overall my todo list became more balanced. 

This way my todo list was also accessible offline, and it was in the same app as my other project-related notes. 

Downsides of Evernote

Unfortunately, there were a few disadvantages as well, that made me miss Todoist. 

The main thing I missed was the integration with Google Calendar – in Todoist I would enter a date and time, and an event would show up on my calendar. Now I had to create a separate “Planning” calendar, and add tasks manually – which I didn’t do consistently.

Another problem was recurrent tasks, which I did once a week or once a month. In Todoist this is basic functionality, but Evernote does not have this feature – you can set a reminder for a note, but when it’s time, you have to reset the reminder yourself. 

Back together

After 2-3 months of using Evernote only, it felt good to create a list in Todoist again. I’m more mindful of the downsides and am trying to manage them better, for example by using filters for my tasks and scheduling tasks for next week on the calendar. It’s not yet the ideal system I wish I would have, but I think using it consistently does help in the long run. 

Do you have any tips of how to create a better todo-list / calendar system? Let me know in the comments!

Firsts: publishing a preprint before submitting the paper

Although I’ve been using preprints since 2013, recently I had a new kind of experience with preprints. Whereas before I would post the preprint upon submitting it to a journal, for the first time I decoupled the two events, and submitted the preprint several months ahead. In this post I reflect on this experience. 

Why

My choice was initially practically motivated. The idea for the paper was born in 2016, but since moving to a new position, I’ve been too overwhelmed to do any writing, so at the start of 2018 the paper – a survey of a magnitude I haven’t attempted before – was far from finished. I wanted a deadline, but I still wanted to be able to update the paper if needed. So a preprint seemed exactly what I needed! 

Timeline

I set the deadline to April 2018 and for the next weeks, worked towards getting the draft to a readable shape.  In April I submitted the preprint to arXiV. Differently from submitting journal-ready preprints, this time I put a piece of text inside the preprint, saying it is not the final version and I was happy to receive suggestions. At the same time, I emailed a few people with the URL asking for comments, and I asked for feedback on Twitter. This felt scary to do – I don’t think I felt as nervous with any of my other papers.

Response

Despite my fears, the experience was positive – I got a lot of constructive feedback which helped me to improve the paper. So in September 2018, I submitted the updated preprint to a journal. In the cover letter, I mentioned the Altmetric statistics of the preprint (I later discovered this is sometimes frowned upon).

Next to the traditional list of suggested reviewers, I also provided several names of people who I had no conflict of interest with, but who had commented on the preprint on their own. I figured that, since they had already read the preprint, they might be willing reviewers. Of course I disclosed this in the letter. 

Publication

The reviews came in about 8 weeks later – an absolute record for me, as during my PhD, regularly waited 6 to 9 months for reviews. The reviewers were constructive, and suggested a coufple of revisions. After revising, the paper was accepted in January 2019 – by that time already gathering a few citations and benefitting from the preprint bump

Verdict

Given this experience, I would definitely post a preprint online without submitting it to a journal first, and not necessarily because of more citations. I realized that me feeling worried about it is a good thing. I could be sure the paper would be seen by a larger group of people, who had an incentive to comment, since they could still influence the paper. This is different from convincing a few reviewers, and then maybe not having the paper noticed afterwards.

Since this paper, I have also been part of another paper that used the same strategy (and is currently under review), and I noticed other preprints putting similar “please email us” messages on the front page. It seems there is a need for interacting with preprints differently – I’m looking forward to what different initiatives like overlay journals will bring.

Not-so-supervised learning of academics

Picking up where I left off with blogging last year – this is part two of a write-up of a talk I’ve given a few times last year, part one is here. After talking about algorithms which deal with data that is not fully labeled, in this part I discuss how career choices can be similar, with my own as an example.

PhD 

Doing a PhD was not on my radar until my MSc supervisor suggested that I apply for a position in the group. I liked the group and doing research for 4 years seemed like a good job to me (see my post on being an employee during your PhD). I didn’t have any specific long-term goal and, as I now realize, was clueless about most aspects of academia. 

What I did understand is that it was good to publish papers. I had a few interesting (though not spectacular) results fairly early on, so I wrote papers and sent them to various workshops. I enjoyed these workshops a lot – since there were not that many people, I could meet researchers I’d just been citing, and have good discussions. On the other hand, I spent quite a lot of time writing smaller papers and pushing away the fact that I needed journal publications to graduate. Also, as I discovered later, my grant reviewers have never heard of these workshops, and thus were not impressed with my publication record. 

I also did a lot of service and outreach activities. I had already been doing this type of thing as a student, so I was good at it, I enjoyed helping others, and it was good for my CV, whether I’d stay in academia or not. So I spent time organizing workshops, reviewing papers, giving talks to encourage more girls into science. I did learn something from all of these activities but in retrospect I think I spent a disproportional amount of time on them. 

Postdoc

I doubted a lot before deciding to go for a postdoc. The awareness of the struggle of finding a position after, and all the people telling me I really have to go abroad to have any shot at it, didn’t help. In the end by talking to more mentors, I decided to go give it a try – without leaving the country. 

My plan was to only do one postdoc and then get an independent position – or leave academia.  As I understood to achieve an independent position I needed to do three things: publish on the project I was hired on, develop my own line of research, and get my own funding. I was not prepared to deal with so many different objectives, so in the end, I did all the things poorly. On top of that, I failed to take care of myself, and had to take a few months off to recover.

It was during a particular low point during the postdoc that I started blogging and tweeting more. It started with me publishing my CV of Failures – I thought I would be documenting a story that would end with me leaving academia. The response was overwhelmingly positive, and I continued with the How I Fail series. During all of this I found an incredibly supportive Twitter community, with many others who were going through similar struggles, and it’s been helpful ever since. 

Tenure Track

Much to my surprise (and other feelings), I did find myself in a tenure track position after all.  This is an important accomplishment, but at the same time, the next goal – getting tenure – is coming up in a few years. Again, there is this (self-imposed?) pressure to do all the things, so it is not without challenges.  But, it is a much better experience in several aspects, because I occasionally realize that I don’t have to do all the things all the time. :

  • I occasionally realize that I don’t have to do all the things all the time. I’ve now actually been able to have periods on time focusing on writing, then focusing on teaching etc. 
  • I occasionally (not often enough) realize that I don’t have to repeat the career paths on others to “succeed”. The combination of things that I do, might just be “good enough”, even though it doesn’t fit the typical “successful” CV. 
  • I have a lot of people, online and offline, who share or have shared many of the same experiences, and who have advice, or are just up for having a coffee or a beer when things are tough.

Academia as supervised learning 

Where does the not-so-supervised learning come in? It seems to me that a lot of advice of what we need to do to “succeed” is based on rules derived from previous “successful” CVs – publishing at particular venues, doing a postdoc abroad, etc. Some of these rules we are explicitly told as advice, others we assume ourselves.

But there is a lot of missing from this picture. The “success” label is a function of much more than particular activities, but also the state of the world (number of tenure track positions, number of students, etc), and the state you are in yourself (including anything else you have to deal with next to the job search). These features have not been taken into account when creating the rules. So even if you do follow all the rules you might get a disappointing outcome, and vice versa.

There might also be opportunities that didn’t exist before. For example, few full professors would have been using Twitter during their PhDs and postdocs. As a yet “unlabeled” activity, it probably wouldn’t come up in any rules, but it can be a powerful tool for early career researchers.

Last but not least, it’s important to remember there’s more than one success metric, and why I’ve been writing “success” in a CV sense. Ultimately success should probably involve being happy, which can be achieved through other types of jobs. And perhaps some of these jobs are not even in our dataset yet.

2019 – year in review

Although I wrote yearly reviews on this blog for several years, I wasn’t expecting to do one this year for two reasons. The first, simple, reason is that I haven’t been blogging recently, and just doing nothing is easier than doing something. The second, more complex reason, is that I might have been afraid to think about this year as a whole. But that is exactly the type of thing that I find important to write about, so here we go.

Mental health

The first thing I have to think about is the manic episode I had in spring which I wrote about earlier, and my diagnosis as bipolar. Mental health issues were not new to me, but this experience was extreme. Although I was stable once I got medication, it felt like parts of my brain had shut off.

Things that were simple before – organizing my todo list, for example – felt completely impossible. I also had let go of many good habits, like running, eating healthy or blogging – pretty much anything I used to write about. I’ve also isolated myself from a lot of people, and felt insecure about most things that I’m normally comfortable with. While my ability to do such things has improved somewhat, more general qualities, like creativity and motivation, did not.

I was only part of the person that I used to be, and this was extremely hard to deal with.

The fact that I am writing this now, probably means that these things are improving too, just at a slower pace. But not feeling this improvement had a huge effect on how I felt this year. Even though a lot of positive things happened, I was often feeling too miserable to properly appreciate them.

Successes

To try to beat that overall feeling, here are a few professional things that went well this year:

  • Received my University Teaching Qualification (a prerequisite for tenure at Dutch universities)
  • Two MSc students graduated!
  • Started supervising two PhD researchers four MSc students (one of whom graduated)
  • My papers on not-so-supervised learning and “Cats or CAT scans” were published and gained a few citations so far (checking Google Scholar way too often)
  • Together with Felienne Hermans, Casper Albers, Natalia Bielczyk and Ionica Smeets, our paper “10 simple rules for starting on Twitter as a scientist” was accepted (online soon!).
  • Together with Natalia Bielczyk, Aidan Budd and Stephan Heunis we got a Mozilla mini-grant and organized a workshop about open & inclusive academia.
  • Visited several places where I gave talks, both on machine learning and topics related to this blog.

Also, an important personal milestone – I got married!

Failures?

When I first started summarizing the positive things I felt guilty. There are many things to be grateful for, but my brain just couldn’t see it that way. In the transition from manic to depressed, I felt bad about many ideas I initiated, but couldn’t follow through on. Afterwards, I felt bad about not doing my part, or not keeping up with my responsibilities. I felt anxious about things I’ve done lots of times.

In retrospect perhaps these things themselves are not failures, the overall failure is that I expected too much of myself. It would have been much better for me to accept how much I’m (not) able to do, let go of everything else, and have patience. Which is why crucial part to this year were the people who experienced me from up close – they were understanding and patient and kind. It’s thanks to them that I’m actually doing alright after what happened, and I’m grateful they are in my life.

Happy new year!

Ups and downs

If you read this blog more often, you might have noticed that it went silent in March 2019. I’ve taken breaks from blogging before, but no break was quite like this, and in this post I explain why.

Although I never wrote about it in detail, I also never made a secret out of the fact that I have been struggling with depression since my postdoc. I had therapy for some of the time and was in general managing things quite well – doing my job, blogging, doing sports, having a social life. The current me almost can’t believe I was able to do all those things. 

In the second half of 2018 things started getting worse. After my cat Buffy passed away in October 2018, I was at an extremely low point and finally decided therapy alone wouldn’t do. My GP prescribed me antidepressants and I started a period of sick leave (full-time at first, part-time later) to adjust.

The antidepressants seemed to be doing an amazing job – the start was slow, but then I started feeling better and better. I soon went back to working full-time, was getting a lot done and had a lot of fresh ideas. I realized I was probably depressed for longer than I thought, and that I was now returning to the “normal” me. This was exciting for me, but somewhat confusing for many people around me, many of whom had not known me that long. 

Eventually – around March – I started feeling a bit too good. The ideas were coming at me so fast I couldn’t keep up, and neither could people interacting with me. My partner recognized this as hypomania, and following a GP visit I was told to stop the antidepressants. The GP also gave me a referral to the psychiatrist, but I ended up on a waiting list. Meanwhile, I was getting more and more out of balance.

The grand finale was a psychotic episode, during which I was convinced that people I’ve never met were giving me clues I had to follow. To top it off, this happened while I was travelling alone. After a few days in a psychiatric facility in France, I was able to return home again, going back on sick leave full-time. The bright side of this episode is that I could see a psychiatrist immediately, who diagnosed me with bipolar disorder.

Now I am getting used to the new medication to stabilize my mood. Although the effects were noticeable straight away and I feel “normal” again, it has been difficult to go back to my regular life with work, blogging, sports, etc, feeling like an impostor in everything. I’m trying to accept that this is normal, and slowly building things up again. I am therefore not sure when the next post might be – but I’ll celebrate that this post is a win.

Not-so-supervised learning of algorithms

About a month ago I gave a talk at UC Dublin, titled “Not-so-supervised learning of algorithms and academics”. I talked both a bit about my research as well as things I’ve learned through Twitter and my blog. The slides are available here but to give some context to everything, I thought I would write a few things about it. In this post I discuss the first part of the talk – not-so-supervised learning of algorithms. All links below are to publishers’ versions, but on my publications page you can find PDFs.

Not-so-supervised learning of algorithms

Machine learning algorithms need examples to learn how to transform inputs (such as images) into outputs (such as categories). For example, an algorithm for detecting abnormalities in a scan, would typically need scans where such abnormalities have been annotated. Such annotated datasets are difficult to obtain. As a result, algorithms could learn input-output patterns that only hold for the training examples, but not for future test data (overfitting).

There are various strategies to address this problem. I have worked on three such strategies:

  • multiple instance learning
  • transfer learning
  • crowdsourcing

Multiple instance learning

The idea in multiple instance learning is to learn from groups of examples. If you see two photos and I tell you “Person A is in both photos”, you should be able to figure out who that person is, even if you don’t know who the other people are. Similarly, we can have scans which have abnormalities somewhere (but we are not sure where), and we can figure out what things they have in common, which we cannot find in healthy scans. During my PhD I worked on such algorithms, and applying them to detecting COPD.

Transfer learning

Another strategy is called transfer learning, where the idea is to learn from a related task. If you are learning to play hockey, perhaps other things you already know, such as playing football, will help you learn. Similarly, we can first train an algorithm on a dataset on a larger source dataset, like scans from a different hospital, and then further train it on our target problem. Even seemingly unrelated tasks, like recognizing cats, can be a good source task for medical data.

There are several relationships between multiple instance learning and transfer learning. To me, it feels like they both constrain what our algorithm can learn, preventing it from overfitting. Multiple instance learning is itself also a type of transfer learning, because we are transferring from the task with global information, to the task with local information. You can read more about these connections here.

Crowdsourcing

A different strategy is to gather more labels using crowdsourcing, where people without specific expertise label images. This has been successful in computer vision for recognizing cats, but there are also some promising results in medical imaging. I have had good results with outlining airways in lung scans and describing visual characteristics of skin lesions. Currently we are looking into whether such visual characteristics can improve the performance of machine learning algorithms.

***

This was an outline of the first part of the talk – stay tuned for the not-so-supervised learning of academics next week!

How to find skirts with pockets

If you have followed my poster skirt story, you might know that my clothes MUST have pockets. Even in 2019, this is somewhat challenging. This causes various problems, including wearing a microphone when speaking.

There are many articles showing how bad the situation is and how we got here. However, I am determined to bring back pockets to women’s clothes.  Here are the strategies I use to still find skirts and dresses with pockets!

Tip 1: Stop buying clothes without pockets

We need to stand together to send a message to clothes manufacturers that we need pockets. It doesn’t matter how cute it is, don’t buy it!


Tip 2: Add pockets to existing clothes

If you did at one point in time buy a skirt or a dress without pockets, consider adding some! I had a pocket added to my dirndl, and it even has a zipper, so I can keep my valuables there safely. 

Me in a dirndl – pocket not visible though 🙁


Tip 3. Search your favorite shop with text queries

Although the websites where I shop (mostly Zalando) allow me to filter very specific criteria, sadly there is no filter for pockets. However, it is possible to do a text search for “steekzakken” (“pockets where you stick things into” in Dutch) to get all products that have this word somewhere. An advantage of this particular term (rather than “zakken” which just means pockets) is that “steekzakken” cannot be used to describe fake pockets that clothes sometimes have. 

Now you get an overview of all clothes with pockets. Browse to dresses or skirts and voila, you have your subset of clothes you can buy! Usually these will all also have a picture (but perhaps not the first picture you see) where the model is using the pockets. This is how I buy most of my skirts and dresses with pockets.   

Tip 4: Use specialized websites

Some brands have caught on to the fact that not having pockets on clothes is stupid. Here are a few – although I haven’t tried any of them yet!

Tip 5: Build an app for it

Building a machine learning algorithm that recognizes “hand in pocket” in pictures should not be too difficult. Add an app on top that searches your favorite websites and gives you all the suggestions you can shop for. Add in ads or sponsorships from companies that do actually put pockets on their clothes, and you have an income stream. I would do it, but I already have too many projects.


Soon after tweeting about this, the idea was picked up by @happybandits, who is now organizing a hackathon for exactly this!  AND, I get to be a pocket consultant! :’)


If you are interested in a pocket hackathon (a #pockethon!), please get in touch with me or @happybandits to see how you can get involved.

Pockets, here we come!

%d bloggers like this: