Abstract: Despite the remarkable achievement of machine learning in various fields, such as music recommendations of Spotify and neural language translation of Google, designing responsible and accountable ML systems remains a significant problem because of the practical requirements of the stakeholders and the regulations in different domains. Interpretable machine learning hereby provides the transparency of the particular decisions made by machine learning systems. The interpretable machine learning tackles the critical problem that humans cannot understand the behaviors of complex machine learning models and how these models arrive at a particular decision. From this perspective, this talk will mainly focus on introducing interpretable machine learning from both theoretical aspects and its applications.
Abstract: Subthalamotomy using transcranial magnetic resonance-guided focused ultrasound (tcMRgFUS) is a novel and promising treatment for Parkinson’s Disease (PD). In this study, we investigate if baseline brain imaging features can be early predictors of tcMRgFUS-subthalamotomy efficacy, as well as which are post-treatment brain changes associated with the clinical outcomes. Towards this aim, functional and structural neuroimaging and extensive clinical data from thirty-five PD patients enrolled in a double-blind tcMRgFUS-subthalamotomy clinical trial were analysanalyzeded. Multivariate cross-correlation analysis revealed that the baseline multi-modal imaging data significantly explain (P<0.005, FWE-corrected) the inter-individual variability in response to treatment. Most predictive features at baseline included neural fluctuations in distributed cortical regions and structural integrity in the putamen and parietal regions. Additionally, a similar multivariate analysis showed that the population variance in clinical improvements is significantly explained (P<0.001, FWE-corrected) by a distributed network of concurrent functional and structural brain changes in frontotemporal, parietal, occipital, and cerebellar regions, as opposed to local changes in very specific brain regions. Overall, our findings reveal specific quantitative brain signatures highly predictive of tcMRgFUS-subthalamotomy responsiveness in PD. The unanticipated weight of a cortical-subcortical-cerebellar subnetwork in defining clinical outcomes extends the current biological understanding of the mechanisms associated with clinical benefits.
Speaker: 顏子昀Zih-Yun Yan, Ph.D. student, New York University
Time: 09/17/2022 05:00 PM PDT 09/17/2022 06:00 PM MDT 09/17/2022 07:00 PM CDT 09/17/2022 08:00 PM EDT 09/18/2022 01:00 AM BST 09/18/2022 02:00 AM CEST 09/18/2022 08:00 AM Taiwan
Field: psychology Sub-field: neuroeconomics
Abstract: Environmental states affect our day-to-day emotions and decisions, including our choices under risk and uncertainty. Past studies have found that an unexpected positive outcome, such as a sunny day after a streak of cloudy days, can increase gambling behaviors. To extend these findings with the broader environmental events to more personal daily life activities at a shorter time scale, we performed longitudinal examinations of the correlation between various daily activities and the change of mood and decision of risk within the individual. Our aim is to identify how engagement in different activities could predict the temporal change in mood and risk preferences. In the talk, I will present data showing how the amount of a particular activity spent last week correlates with the risk attitude-related variables. Among all activities data we had gathered, our finding suggests that subjects have less negative emotion and become more risk-tolerant with more sleep time.