Research

(This site is not currently up to date)

Data Science

Data science has become this buzzword many a people use when they actually mean applying statistics. I am a huge fan of applying statistics, particularly statistical models to gain insights from datasets or to examine how different variables covary (e.g. neural activity during behaviour). I hope to clean up some code and upload it to github soon, and publish some of my kernels on kaggle.  

Publications

Dhole, K. D., Gangal, V., Gehrmann, S., Gupta, A., Li, Z., Mahamood, S., ..., Tolkiehn, M., ... & Zhang, Y. (2021). NL-Augmenter: A Framework for Task-Sensitive Natural Language Augmentation. arXiv preprint arXiv:2112.02721.

Srivastava, A., Rastogi, A., Rao, A., Shoeb, A. A. M., Abid, A., Fisch, A., ..., Tolkiehn, M., ..., & Kim, H. (2022). Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models. arXiv preprint arXiv:2206.04615.

Talks

Tolkiehn, M., (Jan 2018). What can we learn from Telecommunication Data? Talk at Valantic GmbH Hamburg, Germany

Neuroscience - Cerebellar Circuitry

The cerebellum hosts about 5 times as many neurons than the cerebral cortex (21-26 billion neurons in cerebral cortex, Pelvig et al., 2008, in contrast to 101 billion neurons in the cerebellum, Andersen et al., 1992), yet it occupies only about a fifth of the space. It is implied in many aspects of motor control such as smooth movements, yet a complete lack of cerebellum can be in parts compensated for by the rest of the brain. 

I want to understand the cellular and circuit mechanisms of cerebellar activity between its processing stage (cerebral cortex) and its output stage (deep cerebellar nuclei). Then, how does the modulation of network connectivity through various means affect the processing, and how does learning a motor command change the network structure and outputs? 

Neuroscience - Primary Visual Cortex

Neural processing in mouse V1 is part and parcel of modern neuroscience. The majority of research focussed on characterising single cell properties, whilst the role of population activity on processing has only recently been moving into the focus of investigations (Mont.n et al., 2014; Gutnisky et al., 2016; Gardella et al., 2016; O’Donnell et al., 2017; Okun et al., 2012). 

In my work, I aim to analyse and compare individual and population responses (e.g. patterns of co-occurring activity/active variables) with computation models such as maximum entropy models (e.g. Ising model, sRBM) or other statistical or information-theoretic models. Including higher-order interactions or population responses improved the predicted pattern distributions, suggesting the significant role of population dynamics particularly for larger ensemble sizes. 

Publications

Tolkiehn, M., Schultz, S. R. (2015). Multi-Unit Activity contains information about spatial stimulus structure in mouse primary visual cortex. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society

Tolkiehn, M., Schultz, S. R. (2019) Temporo-nasally biased moving grating selectivity in mouse primary visual cortex. Preprint bioRxiv [19 Jul 2019] 10.1101/708644

Tolkiehn, M., Schultz, S. R. (2019). Neural ensemble activity depends on stimulus type in mouse primary visual cortex. Preprint bioRxiv [19 Jul 2019] 10.1101/708636

(more under construction)

Posters (P) and Talks (T)

Abstract: Already in the 1950s it was discovered that moving bars at different orientations elicited responses of varying strength in primary visual cortex. This led to the discovery of orientation-selective neurons (Hubel and Wiesel 1959) and much research into edge detection. However, neural ensemble activity differs between stimulus types: Artificial stimuli such as moving bars or gratings induce responses different from spontaneous activity (no stimulation) or evoked by natural scenes. We found spontaneous and natural scenes-evoked activities indicated lower firing rates, Shannon entropies, and binary word distribution divergences (Jensen-Shannon-Divergence) than either to drifting gratings.

Ethomics

Ethomics is the neologism of the "-omics" suffix (which describes a ) and ethology (the study of human/animal behaviour from a biological perspective)

Ethomics looks at the quantification of behaviour. Similar to techniques in body sensor networks, I was interested in observing and quantifying the normal behavioural movement types in healthy persons using a full-body motion capturing system (with 48 sensors streaming 3d accelerometer data) in activities of daily living scenarios (e.g. sleeping, eating breakfast, working at a computer). The constitution of a database of unconstained natural movements then allows us to objectively quantify and contrast healthy movements with those of Parkinsons patients. In our studies, we found differences in the complexity of movements across genders and health state. 

Poster and presentations

Thomik, A., Tolkiehn, M., Vella, I. and Faisal, A. A. (2012), Human ethomics as a window into motor control, Champalimaud Neuroscience Symposium, Lisbon, Portugal

Tolkiehn*, M., Vella*, I., Thomik, A. and Faisal, A. A. (2012), Towards the Human Ethome: Collection and Analysis of Unconstrained Natural Human Movements, Bioengineering12, Oxford

Ubiquitous sensing/Body Sensor Networks

Ubiquitous sensing deals with highly portable and without a problem wearable sensory devices. 

A personal experience of my grandmother falling over in the garden without anyone noticing inspired me to delve into researching fall detection for the elderly. I devised and tested an algorithm that reliably detects and classifies falls from non-falls, simple enough to be ported onto a microcontroller of a triaxial accelerometer and barometric pressure sensor of a small waist-worn device. 

Publications

Tolkiehn, M., Atallah, L., Lo, B. and Yang, G.-Z. (2011) Direction-sensitive fall detection using a triaxial accelerometer and a barometric pressure sensor. Proceedings of the 33rd Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS