Papers for the month of January 2020

Plaut, David C.

"Hemispheric organization for visual object recognition: A theoretical account and empirical evidence "
Perception, TBA:TBA

Mouse over here for a brief summary or click to open article in a new tab.http://Despite the similarity in structure, the hemispheres of the human brain have somewhat different functions. A traditional view of hemispheric organization asserts that there are independent and largely lateralized domain-specific regions in ventral occipitotemporal (VOTC), specialized for the recognition of distinct classes of objects. Here, we offer an alternative account of the organization of the hemispheres, with a specific focus on face and word recognition. This alternative account relies on three computational principles: distributed representations and knowledge, cooperation and competition between representations, and topography and proximity. The crux is that visual recognition result from a network of regions with graded functional specialization and distributed across both hemispheres. Specifically, the claim is that face recognition, which undergoes acquisition early in life, is processed by VOTC regions in both hemispheres. Once literacy is acquired, word recognition, which is co-lateralized with language areas, primarily engages the left VOTC and, consequently, face recognition is primarily, albeit not exclusively, mediated by the right VOTC. We review psychological and neural evidence from a range of studies conducted with normal and brain-damaged adults and children and consider findings which challenge this account. Last, we offer suggestions for future investigations whose findings may further refine this account.

Tervo-Clemmens, B., Quach, A., Calabro, F.J., Foran, W.,

"Meta-analysis and review of functional neuroimaging differences underlying adolescent vulnerability to substance use"
NeuroImage, 209:116476

Mouse over here for a brief summary or click to open article in a new tab.Adolescence is increasingly viewed as a sensitive period in the development of substance use disorders (SUDs). Neurodevelopmental ‘dual-risk’ theories suggest adolescent vulnerability to problematic substance use is driven by an overactive reward drive mediated by the striatum, and poor cognitive control mediated by the prefrontal cortex. To this end, there has been a growing number of neuroimaging studies examining cognitive and affective neural systems during adolescence for markers of vulnerability to problematic substance use. Here, we perform a coordinate-based meta-analysis on this emerging literature. Twenty-two task-based voxelwise fMRI studies with activation differences associated with substance use vulnerability, representative of approximately 1092 subjects, were identified through a systematic literature search (PubMed, Scopus) and coordinates of activation differences (N ​= ​190) were extracted. Adolescents were defined as ‘at-risk’ for problematic substance use based on a family history of SUD or through prospective prediction of substance use initiation or escalation. Multilevel kernel density analysis was used to identify the most consistent brain regions associated with adolescent substance use vulnerability. Across the included studies, substance use vulnerability was most reliably associated with activation differences in the striatum, where at-risk adolescents had hyper-activation in the dorsal subdivision (putamen). Follow-up analyses suggested striatal differences were driven by tasks sharing a motivational and/or reward component (e.g., monetary incentive) and common across subgroups of substance use risk (family history and prospective prediction studies). Analyses examining the role of psychiatric comorbidity revealed striatal activation differences were significantly more common in samples whose definition of substance use risk included cooccurring externalizing psychopathology. Furthermore, substance use risk meta-analytic results were no longer significant when excluding these studies, although this may reflect limitations in statistical power. No significant activation differences were observed in prefrontal cortex in any analysis. These results suggest striatal dysfunction, rather than prefrontal, may be a more primary neural feature of adolescent vulnerability to problematic substance use, possibly through a dimension of individual variability shared with externalizing psychopathology. However, our systematic literature search confirms this is still an emerging field. More studies, increased data sharing, and further quantitative integration are necessary for a comprehensive understanding of the neuroimaging markers of adolescent substance use risk.

Witherspoon, E.

"Locating and understanding the largest gender difference in pathways to science degrees"
Science Education, 104:144-163

Mouse over here for a brief summary or click to open article in a new tab.Studies addressing gender inequity in science often ignore a large source of undergraduate science degree earners: those who enroll in science courses intending to pursue careers in health or medicine. This study examines pathways toward or away from science degrees in N = 4,345 men and women enrolled in early science courses at a large undergraduate research university.

Toneva, M, and Wehbe, L.

"Interpreting and improving natural-language processing (in machines) with natural language-processing (in the brain)"
Advances in Neural Information Processing Systems, 33:14928-14938

Mouse over here for a brief summary or click to open article in a new tab.Neural networks models for NLP are typically implemented without the explicit encoding of language rules and yet they are able to break one performance record after another. This has generated a lot of research interest in interpreting the representations learned by these networks. We propose here a novel interpretation approach that relies on the only processing system we have that does understand language: the human brain. We use brain imaging recordings of subjects reading complex natural text to interpret word and sequence embeddings from 4 recent NLP models - ELMo, USE, BERT and Transformer-XL. We study how their representations differ across layer depth, context length, and attention type. Our results reveal differences in the context-related representations across these models. Further, in the transformer models, we find an interaction between layer depth and context length, and between layer depth and attention type. We finally hypothesize that altering BERT to better align with brain recordings would enable it to also better understand language. Probing the altered BERT using syntactic NLP tasks reveals that the model with increased brain-alignment outperforms the original model. Cognitive neuroscientists have already begun using NLP networks to study the brain, and this work closes the loop to allow the interaction between NLP and cognitive neuroscience to be a true cross-pollination.

Tarr, M., Wang, A.,

"Neural Taskonomy: Inferring the Similarity of Task-Derived Representations from Brain Activity"
Advances in Neural Information Processing Systems, 32:15475-15485

Mouse over here for a brief summary or click to open article in a new tab.This paper helps reveal the task-specific architecture of the human visual system.

Schwartz, D., Toneva, M., Wehbe, L.

"Inducing brain-relevant bias in natural language processing models"
Advances in Neural Information Processing Systems (NeurIPS), 2019:14100-14110

Mouse over here for a brief summary or click to open article in a new tab.We demonstrate that a powerful pre-trained language model recently introduced by the natural language processing community (BERT) can be fine-tuned to predict functional magnetic resonance imaging (fMRI) data from participants reading a story. We also show that the model learned by BERT during this fine-tuning transfers across multiple participants, and that, for some participants, the fine-tuned representations learned from both magnetoencephalography (MEG) and fMRI are better for predicting fMRI than the representations learned from fMRI alone, indicating that the learned representations capture brain-activity-relevant information that is not simply an artifact of the modality.


  • 2019
  • 2018
  • 2017
  • 2016
  • 2015
  • 2014
  • 2013