Research Aims

My research takes a deeply interdisciplinary approach to the study of human behavior, drawing insights and tools from clinical psychology, computer science, and quantitative methods. Each of these fields offers a unique perspective on human behavior and the problems that arise during its examination. I have organized my early career around the notion that the advancement of scientific understanding will require individuals capable of traversing the borderland where these fields meet. In addition to serving as an interdisciplinary liaison and translator, I strive to integrate these perspectives into more nuanced and detailed wholes.


Clinical Psychology

I am primarily interested in discovering how personality and psychopathology manifest in observable affective and interpersonal behaviors, especially in the context of dyadic and small group interactions. Toward this end, I have studied how various psychological disorders (e.g., depression and personality disorder) affect the ways that patients interact with significant others in their lives, such as clinicians, romantic partners, and family members (e.g., parents and children).

To date, my contributions to clinical psychology include the popularization of automated tools for the analysis of depression (Girard & Cohn, 2015) and the “Social Withdrawal” hypothesis: that interpersonal behavior in depression often facilitates social withdrawal and increases interpersonal distance during an interaction (Girard et al., 2013, 2014), rather than (or in addition to) signaling negative affect or psychomotor retardation. Work currently under review and in preparation extends my work on depression and also applies similar methodologies to other conditions, such as personality pathology and post-traumatic stress.

Computer Science

I am also interested in leveraging computer vision, machine learning, and software engineering for the advancement of psychological research, especially the analysis of human behavior. Toward this end, I have developed algorithms for the automatic analysis of facial expressions, methods to validate the performance and generalizability of such algorithms, and software to assist in the annotation of multimedia by human raters (for psychological study and algorithm training).

To date, my contributions to computer science include novel methods for developing and validating supervised learning algorithms and a psychology-focused perspective on the automatic analysis of human behavior. One recent focus of my work has been to push for explicit analysis of the intensity of facial expressions, rather than just their presence or absence (Girard, Cohn, & De la Torre, 2015; Jeni, Girard, Cohn, & De la Torre, 2013; Valstar, Almaev, Girard, et al., 2015). Another focus has been to model the factors that influence an algorithm’s performance, including characteristics of the data (Girard, Cohn, Jeni, Lucey, & De la Torre, 2015; Girard, Cohn, Jeni, Sayette, & De la Torre, 2015) and assumptions of the metrics used to quantify performance (Girard & Cohn, 2011).


Quantitative Methods

I am also interested in developing, using, and popularizing new methods for research design and data analysis. Toward this end, I have begun developing expertise in advanced statistical techniques including multilevel modeling, meta-analysis and meta-regression, missing data analysis, inter-observer reliability analysis, and structural equation modeling. Many of these techniques are poorly known in the computer science and affective computing fields and my work has pushed for greater awareness of the importance of acknowledging and addressing issues such as data interdependence, missing data, and chance agreement.

To date, my contributions to quantitative methods have been modest. However, as stated previously, my work has brought awareness in the affective computing field to several key statistical issues (e.g., Girard & Cohn, 2011; Girard, et al., 2015). Work currently under review and in preparation extends this work to additional techniques and also develops a novel approach to estimating inter-observer reliability.