Sohrob Saeb

Postdoctoral Fellow
Department of Preventive Medicine, Northwestern University

s-saeb aht northwestern doht edu
Google Scholar


I received my PhD in Computer Science (Computational Neuroscience) from Goethe-Universität Frankfurt, Germany, focusing on modeling the neurocomputational mechanisms of coordinated eye and head movements in primate brain. I used artificial neural networks and reinforcement learning to model adaptive control of saccades and to deal with uncertainty in active vision. My current research at Northwestern University focuses on the application of machine learning and sensor technology to mental healthcare, where I use the sensor data collected from smartphones and wearable devices to infer information about different aspects of individuals’ daily-life behaviors and their environments (see Current Projects).

Current Projects

Life Sensing

Depression and anxiety are among major health concerns and a growing problem in the modern society. Smartphones and wearable devices, with the variety of sensors (e.g. GPS, accelerometer, and microphone) and other sources of information (e.g., call, sms, and games) can potentially act as bio-markers of depression and anxiety in daily life, providing an opportinuty to more closely investigate the life-long trajectory of these illnesses.

The goal of this project is to investigate if smartphone sensor data can be used to (1) distinguish between depressed and non-depressed individuals, (2) predict the severity of the disease, and (3) predict the future trajectory that the disease takes in an individual. We have been able to do this using GPS data in a small study, which was successfully replicated on an independent dataset StudentLife. We are now collecting data from 240 individuals to extend our findings to other aspects of life including sleep, social interactions, and physical activity.

Adversarial Activity Recognition

Current smartphone and wearable sensor technologies allow us to detect a person’s physical activities such as walking, biking, or being still. In many circumstances, however, there are strong incentives for users to trick the activity trackers into detecting activities other than the ones they actually perform. For example, one can make their pedometer count steps by shaking their device (many more ways shown here). In fact, current activity recognition technology is only reliable in normal conditions, and thus vulnerable to deceptive behavior. The aim of this project is to develop a methodology that enables smartphone-based activity recognition to overcome this limitation.

We have already conducted a study, in which we asked volunteers to deliberately make the activity tracker fail, and then used those data to re-adapt the tracker. We aim to extend our methodology to more challenging tasks as well as other modalities in future.

Cross-Validation in Clinical Predictions

The application of machine learning in clinical predictions is rapidly growing. Examples range from using accelerometers to detect the severity of Parkinson’s disease to analyzing speech for predicting the onset of psychosis. A crucial step in using machine learning is to properly evaluate their accuracy, and this is performed through various cross-validation methods.

In a recent study (preprint), we showed how a popular cross-validation method massively overestimates the prediction accuracy of the algorithms. We also performed a systematic review of the literature, and showed that about 45% of the studies using machine learning for clinical predictions have used the wrong method. The results will be published in GigaScience.



S Saeb, EG Lattie, K Körding, DC Mohr, Semantic location from mobile phones: Going beyond GPS. under review.

S Saeb, T Cybulski, K Körding, DC Mohr, Scalable passive sleep monitoring using smartphones: opportunities and obstacles. Journal of Medical Internet Research, 19(4):e118, 2017.

MA Little, G Varoquaux, S Saeb, L Lonini, A Jayaraman, DC Mohr, K Körding, Using and understanding cross-validation strategies. Perspectives on Saeb et al.. GigaScience, 2017.

S Saeb*, L Lonini*, A Jayaraman, DC Mohr, K Körding, The need to approximate the use-case in clinical machine learning. GigaScience, 2017.

S Saeb*, L Lonini*, A Jayaraman, DC Mohr, K Körding, Voodoo machine learning for clinical predictions. bioRxiv, 059774, 2016.

S Saeb, E Lattie, SM Schueller, K Körding, DC Mohr, The relationship between mobile phone location sensor data and depressive symptom severity. PeerJ 4:e2537, 2016.

S Saeb, K Körding, DC Mohr, Making activity recognition robust against deceptive behavior. PLoS ONE 10(12), 2015, e0144795.

S Saeb, M Zhang, CJ Karr, SM Schueller, ME Corden, K Körding, DC Mohr, Mobile phone sensor correlates of depressive symptom severity in daily-life behavior: an exploratory study. Journal of Medical Internet Research, 17(7), 2015, e175.

S Saeb, C Weber, J Triesch, Learning the optimal control of coordinated eye and head movements. PLoS Computational Biology, 7(11), 2011, e1002253.

S Saeb, C Weber, J Triesch, Goal directed learning of features and forward models. Neural Networks, 22 (5-6), 2009, pp. 586-592. PDF

A Farajidavar, S Saeb, K Behbehani, Incorporating spike-timing dependent plasticity (STDP) and dynamic synapse into a computational model of wind-up. Neural Networks, 21(2-3), 2008, pp. 241-249.

M Amiri, S Saeb, MJ Yazdanpanah, SA Seyyedsalehi, Analysis of the dynamical behavior of a feedback auto-associative memory. Neurocomputing 71(4-6), 2008, pp. 486-94.

S Saeb, S Gharibzadeh, F Towhidkhah, A Farajidavar, Modeling the primary auditory cortex using dynamic synapses: can synaptic plasticity explain the temporal tuning? Theoretical Biology, 241(1), 2007, pp. 1-9.

A Farajidavar, S Gharibzadeh, F Towhidkhah, S Saeb, A cybernetic view on wind-up. Medical Hypotheses, 67(2), 2006, pp. 304-6.

Conference Papers

S Saeb, M Zhang, MM Kwasny, CJ Karr, K Körding, DC Mohr, The Relationship between Clinical, Momentary, and Sensor-based Assessment of Depression, In 9th International Conference on Pervasive Computing Technologies for Healthcare (Pervasive Health), 2015. PDF

S Saeb, C Weber, J Triesch, A Neural Model for the Adaptive Control of Saccadic Eye Movements, International Joint Conference on Neural Networks (IJCNN), Atlanta, Georgia, USA, 2009, pp. 2740-2747. PDF

S Saeb, A Farajidavar, S Gharibzadeh, A model of wind-up based on short-term and long-term synaptic plasticity mechanisms, International Joint Conference on Neural Networks (IJCNN), Orlando, Florida, USA, 2007, pp. 1055-1060. PDF

S Saeb, M Amiri, MJ Yazdanpanah, Analysis of the dynamical behavior of a feedback auto-associative memory, IEEE International Symposium on Neural Networks (ISNN), Nanjing, China, 2007. PDF

Conference Talks/Posters

S Saeb, CJ Karr, SM Schueller, DC Mohr, Passive detection of depression using features of GPS location: two studies, International Society for Research on Internet Interventions (ISRII) Scientific Meeting, Seattle, Washington, USA, 7-9 April 2016.

S Saeb, K Körding, DC Mohr, Human vs. machine: Improving physical activity tracking in the presence of deceptive human behavior, Society for Neuroscience (SfN), Chicago, IL, USA, 2015. abstract

S Saeb, C Weber, J Triesch, Development of coordinated eye and head movements during gaze shifts, Vision Sciences Society Annual Meeting, Naples, Florida, USA, 2011. poster

S Saeb, C Weber, J Triesch, Learning coordinated eye and head movements: unifying principles and architectures. Frontiers in Computational Neuroscience, Conference Abstract: Bernstein Conference on Computational Neuroscience, Berlin, Germany, 27 Sep - 1 Oct, 2010. abstract, talk

S Saeb, C Weber. Toward a goal-directed construction of state spaces. Frontiers in Computational Neuroscience, Conference Abstract: Bernstein Conference on Computational Neuroscience, Frankfurt am Main, Germany, 30 Sep - 2 Oct, 2009. abstract, poster

S Saeb, A Farajidavar, Spike-time dependent plasticity could explain temporal tuning of auditory cortical cells, Computational and Systems Neuroscience (COSYNE), Salt Lake City, Utah, USA, 2008. abstract


     Activity trackers could be better. So why aren’t they? January 2016.

The fitness tracker that knows if you’re faking: Researchers train smartphones to spot cheaters. January 2016.

Smarter activity tracker knows when you’re just pretending to work out January 2016.

Your phone knows if you’re depressed. July 2015.

Can your smartphone tell you if you have depression? July 2015.

Can your phone really know you’re depressed? July 2015.

      Northwestern study: Your smartphone knows if you’re depressed July 2015.

Your phone can tell whether you’re depressed. July 2015.