A Call to Attention: How Deep Learning can Improve our Understanding of OCD

Gretchen DobbinSchmaltz

University of Houston

A particularly mystifying feature of Obsessive-compulsive Disorder (OCD) is that obsessions, characteristic of the disorder, can trigger anxiety despite the ability of OC individuals to recognize such intrusive mental content as irrational. The framing of OCD as a disorder of attention provides some insight as to how impaired cognitive processes give rise to powerful effects of obsessions like intrusive thoughts. Psychological research suggests that, as a result of innate deficits in selective attention, OC individuals struggle to dis-attend to irrelevant cognitive and sensory stimuli (Clayton et al., 1999). This may offer insight as to why intrusive thoughts register as salient features of the mental stream in OCD. Additional research shows that, for OC individuals, anxiety triggered by intrusive mental content negatively impacts performance on OCD-unrelated tasks requiring selective attention (Cohen et al., 2003).  Understanding how selective attention functions in the OC brain is critical to an ultimate understanding of how intrusive mental content prompts obsessive-compulsive behavior.                                                         

One way to deepen our understanding of selective attention in OCD is to explore artificial models of attention within deep learning. With psychological theories of selective attention for OCD in-hand, there exists an exciting opportunity for synergy between research in OCD and artificial intelligence. In this work, I explore how experimental results of Kaplan et al. (2005), in which OC individuals demonstrate potentiated latent inhibition effects, might be used as a target for performance on similar visual search tasks in DNNs. If psychological theories about how selective attention links to this effect are correct, DNNs with artificial attention mechanisms should be capable of producing similar results under adjusted parameters.                                                        The approach serves not only to assess and refine psychological theories of selective attention in OCD, but to facilitate progress toward developing biologically plausible models of attention in DNNs.  By testing psychological theories of attention through adjustment of independent variables, the strategy addresses the work of Bowers et al. (2005), who argue that current benchmarks for bioplausibility are inadequate to ground claims that some DNNs are apt models for biological systems.

Chair: Sharon Casu

Time: September 6th, 18:50-19:20

Location: SR 1.003