i_mBODY lab

Date: 2024-09-25, 12:00h-13:00h UTC+2 (CEST)

Speaker: Freek Hens, Radboud University and the Australian National University

Title: From continuous biosignals to spike trains: LAST opens the door to the effective training of spiking neural networks on continuous signals

Abstract: 

Can we use the brain’s own language to diagnose chronic lower back pain? In this talk, I’ll present our groundbreaking work applying Spiking Neural Networks (SNNs) to classify chronic lower back pain using the EmoPain dataset. This is the first time SNNs have been used for this purpose.

We’ve developed a novel encoding technique called Learning Adaptive Spike Thresholds (LAST) to efficiently convert continuous biosignals into spike trains. Combined with an ensemble of Spiking Recurrent Neural Networks (SRNNs), our approach effectively handles the complexities of multi-stream sEMG and IMU data.

Despite the challenges of limited data and imbalanced classes, we’ve achieved impressive results, outperforming traditional methods and even state-of-the-art deep learning models. Our findings pave the way for energy-efficient, wearable solutions for chronic pain management. Join me to learn more about our innovative approach and its potential impact.

Biography:

Freek Hens holds Bachelor’s and Master of Science degrees in AI from Radboud University, Nijmegen, the Netherlands. He is currently pursuing additional studies at the Australian National University (ANU) in Canberra, Australia. His research interests span efficient neuromorphic computing, particularly Spiking Neural Networks (SNNs), Computer Vision, and Natural Language Processing (NLP).

His Master’s thesis focused on developing energy and time-efficient AI models using SNNs, resulting in a cum laude distinction and a forthcoming journal publication. To broaden his AI expertise, he is exploring Computer Vision at ANU. Additionally, Freek gained experience in NLP during an internship centred on large-scale text classification with limited labelled data.

Date: 2024-05-28, 11:00h UTC+2

Speaker: Mohammad (MH) Rahmani, PhD research candidate in Computer Science and Technology at UC3M

Title: A journey to re-embodiment

Abstract:

In this presentation, I will discuss my past, highlighting the relevant experiences and projects that have brought me here today. I will go into parts of my portfolio that showcase my abilities, as well as my thought process and general interests, both historically and moving forward. This will hopefully give you a good picture of what you can expect from me and how I can be helpful in the future.

Biography:

Mohammad (MH) Rahmani is a PhD research candidate in Computer Science and Technology at UC3M, currently focusing on Embodied Perception through Interaction and Sensations.

He was a research fellow at SAIL Lab in University of New Haven focusing on real-time neurofeedback systems, signal processing, and artificial intelligence in Brain-Computer Interfaces (BCI); as well as developing Text-to-Speech (TTS) models for low-resource languages. With a background in observational astronomy, computer science, and multi-media production, Rahmani holds a Master’s degree from NYU’s Interactive Telecommunications Program (ITP) and a BSc in Computer Science. He has professionally worked in R&D roles on LoRa Chirp based IoT consumer devices, safety surveillance systems, and image processing.

Additionally, he has served as the Vice Chair of Tehran ACM SIGGRAPH, and his creative work has been featured in various galleries, hackathons, and festivals. He is an interactive media artist and new media researcher known for immersive spatial mixed reality experiences. His work has been showcased at Archuwum Art Transparent, TADAEX Festival, SIGGRAPH Conference, and SFA Projects.

Date: 2024-05-07, 14:00h UTC+2

Speaker: Miguel A. Vadillo, Associate Professor at Universidad Autónoma de Madrid, Director of the Cognition, Attention, and Learning Lab (CALL)

Title: Seeing (a pre-registered protocol) is believing

Abstract: 
How much confidence should you put on a particular hypothesis if p < .05? From a rational point of view, the credibility of a scientific finding depends on three parameters: a) the prior likelihood of the hypothesis being true, b) the probability of a significant result if the alternative hypothesis is true, and c) the probability of a significant result if the null hypothesis is true. Unfortunately, (a) is largely subjective. (b) and (c) cannot be properly estimated with the information typically reported in scientific papers, but meta-scientific studies suggest that there are reasons to be pessimistic. Psychological research is often underpowered, which means that (b) is too low, and everything suggests that many significant findings could be the result of questionable research practices, casting doubt on (c). When we read a non-registered study, we can never know if the result is reliable or if the data were simply “tortured until they confessed”. The crucial contribution of pre-registration is that it ensures that the false positive rate is not inflated by questionable research practices or selective reporting. Furthermore, pre-registration forces researchers to plan their sample size based on a formal power analysis, ensuring that (b) is sufficiently large. As a result, in this talk I will defend that it is logical to put more faith in pre-registered studies than in non-registered research. Although non-registered research is not without value, pre-registration is definitely worthwhile.

Biography:
Miguel Vadillo is associate professor at Universidad Autónoma de Madrid, where he coordinates the Cognition, Attention, and Learning Lab (CALL). He is also the current president of the Spanish Society for Experimental Psychology, member of the ethical review board at UAM, and member of the scientific committee of the National Research Agency. His research is focused on statistical problems in research on implicit learning and other unconscious mental processes and also on meta-analytic methods for the detection of biases in scientific research.
https://mvadillo.com/