
If you work in digital mental health or wearable tech and would like to grab a coffee in London or chat online, feel free to reach out!

Psychiatrist | IPPRF Fellow @imperialcollege - Digital Mental Health

If you work in digital mental health or wearable tech and would like to grab a coffee in London or chat online, feel free to reach out!

I’ve had a wonderful five years in Edinburgh. I am grateful to everyone I met while working as a psychiatrist for NHS Lothian and during my PhD at University of Edinburgh. I look forward to building on the skills and experiences I gained there as I begin this next chapter.

I've recently joined Imperial College London as an IPPRF Fellow in Psychiatry in the Department of Brain Sciences. I will be working on AI- and wearable technology-based approaches to improve care for people with mood disorders and deepen our understanding of these conditions.

3/3 Collecting HRV data and psychometric scales over multiple time points during a BD episode is costly, limiting the sample size. We introduce a Bayesian Hierarchical Model, better suited for small samples and uncertainty quantification.

2/3 Heart rate variability (HRV) - variability in the time between consecutive heartbeats -reflects the health of the autonomic nervous system. Our study shows that HRV recovery correlates with symptom improvement in BD 🎭, suggesting HRV could be a potential biomarker for BD.

1/3 Excited to share our tell-tale heart study on Bipolar Disorder (BD), now published in @npjMentalHealth : nature.com/articles/s44184-0… @tetraduzione @bryanlimy @DHidalgoMazzei @BioMedAI_CDT @BipolarBCN

and congratulations to Dr. @filippocmc who successfully passed his viva, examined by Paolo Ossola and @KiaNazarpour Filippo's thesis is a wonderful bridge between #ML #AI and clinical practice in psychiatry for mood disorders with wearable!
1/2 If you are working in #wearables, #AI, and #healthcare do not let small datasets stop you. mhealth.jmir.org/2024/1/e550… 👈 We share the largest publicly available data collection for #empatica #e4 and release the codebase for pre-processing and self-supervised pre-training.

2/2 Repurposing unlabelled data with self-supervised learning can help with your task of interest. @tetraduzione @DHidalgoMazzei @bryanlimy @BioMedAI_CDT @ancAtEd Data: huggingface.co/datasets/FcmC…. Codebase: github.com/april-tools/e4sel….

1/2 If you are working in #wearables, #AI, and #healthcare do not let small datasets stop you. mhealth.jmir.org/2024/1/e550… 👈 We share the largest publicly available data collection for #empatica #e4 and release the codebase for pre-processing and self-supervised pre-training.

#NeuroSymbolic #AI 🤖 enforces constraints, but models can achieve high accuracy using wrong concepts. Can we spot when a model relies on flawed concepts and ensure #trustworthiness? Yes, with BEARS! 🐻 📢 Introducing our latest paper, accepted as a spotlight at UAI2024 !🎉📄

we evaluate #continual learning models on "baby" benchmarks, where it is easy to show no catastrophic forgetting! we propose a new simple benchmark that glues simple but still challenging tasks in a curriculum: from MNIST to Imagenet and back! 📜link.springer.com/article/10…

Will be at AISTATS this week, would love to chat if you have integrals to approximate or are generally into compstat / opt. transport / estimation in causal inference. Also we have two papers around importance sampling and variational inference. 1st: adaptive IS for heavy tails

The softmax bottleneck is an interesting problem; it has many side effects which we do not yet fully understand! If you want to build an intuition for the problem, here is an interactive visualisation I made grv.unargmaxable.ai/static/f… (best viewed on desktop).
(6/10) This problem is in fact very much related with the softmax bottleneck issue (arxiv.org/abs/1711.03953) Basically, we try to map "low" dimensional contextual representations to potentially high-dimensional contextual probability manifolds, using a simple linear layer:

New preprint out!🚨We performed a large scale analysis of physicochemical features extracted from over half a million AlphaFold structural models, and using data-driven methods we showed a link between these features and in vivo behaviour of proteins. Find out more below⬇️1/9
Large scale analysis of predicted protein structures links model features to in vivo behaviour biorxiv.org/cgi/content/shor… #bioRxiv


Classical mixture models are limited to positive weights and this requires learning very large mixtures! Can we learn (deep) mixtures with negative weights? Answer in our #ICLR2024 spotlight by @loreloc_ Aleks, Martin, Stefan, Nicolas @arnosolin 📜openreview.net/forum?id=xIHi…
After the score-based models, I decided to take a step back and cover the basics in a new blog post: Probabilistic modeling and Mixture Models. Additionally, I make a brief intro to ✨Probabilistic Circuits✨ Check: 📄Post: jmtomczak.github.io/blog/19/… 🖥️ Code: github.com/jmtomczak/intro_d…