Writing this report was one requirement of the travel grant which let me attend this conference. Having written and hyperlinked it up, I thought I'd post it in case it would be of interest to anyone. At least check out the videos!
2016 was the second time that I visited the Annual Meeting of the Organization for Human Brain Mapping, this time to present the poster 3682: Investigating human speech recognition: Reverse-engineering the machine solution with EMEG and RSA, co-authored by my colleagues at the Departments of Psychology and Engineering at Cambridge. The visit was generously supported by a Brain travel grant, for which I give my thanks.
Arriving on Sunday 26th, I was able to attend the Daniel Wolpert opening lecture on reverse-engineering the brain’s algorithms for sensorimotor control. (Check out this interview of Wolpert by Niko Kriegeskorte.) This was followed by the welcomming drinks reception, with the main events and sessions starting the following day.
The first may day (27th) opened with a symposium addressing the disentangling of association, correlation and causation in neuroimaging studies. Of particular interest to me was Sebastian Weichwald’s session on obtaining causal hypotheses from neuroimaging studies, accounting for hidden variables, and the interpretative differences between encoding and decoding models. A useful follow-up to this was given by Weichwald’s paper on the same topic, in NeuroImage (preprint on ArXiv).
The symposium on disorders of self-awareness and dissociative states, while not directly relevant to my current research, was nonetheless fascinating and inspiring. Being exposed to such things is important for someone like me who is transitioning into neuroscience from another field. Stand-out sessions were Olaf Blanke’s on bodily self-consciousness, a phenomenon which is disrupted by illusions such as the rubber hand illusion and its whole-body equivalent, which cause a person to wrongly identify with an avatar rather than their real body.
Also Melanie Boly’s talk on quantifying consciousness, highlighting how difficult it can be behaviourally, given that consciousness is not determinable by responsiveness (as evidenced by locked-in syndrome and REM sleep). However, a study she presented showed that EEG taken during sleep could strongly predict the presence of conscious experience.
Posters of particular interest to me on this day (all posters and abstracts available here, but direct linking is not possible): 1731: Neural correlates of auditory artificial grammar learning, which showed differences in activity in premotor and dorsal areas in subject groups who performed well at learning an abstract phonetic grammar. 1661: The “Brain Program” of the Programmer’s Brain: is it Language or Math that matters?, which showed that mathematics-associated areas were employed by programmers while bugfinding and mentally tracking recursion, whereas linguistic areas were used for general understanding of source code. As well as being a programmer and mathematician myself, I am particularly interested in the differences in neural processing of formal and informal grammars.
Haiguang Wen’s session in the perceptual representations oral session was remarkable — using a deep learning framework to decode and categorise images from fMRI data recorded from subjects watching natural movies. This was a really excellently conducted and presented study. Though an emergent theme to me on researchers using neural network models was the loose use of the word “semantic” to mean essentially “object identity”; cf. work by Barry Devereux which unpacks different meanings of “semantics” for visual object recognition. Jean-Remi King’s talk in this session also presented some very nice visualisation techniques.
Posters I saw on the next day which stood out: 1505: Where the rhythm plays: Machine learning decodes rhythm-sensitive cortices, decoding heard rhythm identity as distinct from instrument identity using a SVM technique. 2279: Contextual effects on the neural encoding of speech sounds, interesting study showing different features (low versus high frequency) used by listeners in different tasks (speaker vs phoneme detection).
Niko Kriegeskorte’s session in the symposium on “revealing fine-scale representation in human cortex” had new and interesting things to say about simulating voxel size when interpreting fMRI RSA results. Important and subtle points which seem to be often overlooked in such studies.
The symposium on neural networks and deep learning on the morning of the 29th opened with a good introduction to the subject by Ahmed Abdulkadir, including a section on DNN-based automatic image segmentation with projects like DeepMedic.
Later, Susan Bookheimer’s eye-opening keynote on the hippocampus and medial temporal lobe memory loss was excellent and highly enjoyable.
This was followed by an outstanding keynote by Tim Behrens on global versus local encoding of world models. This featured his work showing the existance of grid-cell-like behaviour encoding abstract two-dimensional spaces. A great and inspiring talk!
On the final day (30th), I saw a number of good posters. 3678: How much cortex do we need to decode speech sounds? on the minimum amount of cortex needed to decode speech sounds was very interesting, and findings somewhat different to those on my own poster! I have contacted the author to find out more; the study seemed well-conceived and conducted, and worth understanding in detail. 3996: Multivariate distance correlation is a more reliable and robust measure of functional connectivity proposed the distance correlation as a more informative alternative to the ubiquitous Pearson’s correlation used in RSA and other multivariate pattern-information analyses. Another author I have made contact with, as I want to fully understand the mathematics of these claims.
Overall, an extremely enjoyable and edifying conference experience. I am pleased to say that my own poster seemed to be of interest to several people, and I had a busy presentation session. I have already had a number of follow-up enquiries since.
I am again very grateful to the Guarantors of Brain for financially supporting my attendance.