Research Lines

1. Visual Perception

I am interested in how the human visual system extracts information about the physical world to guide action.

My approach: Scaling principled, interpretable models to develop mechanistic understanding.

Mechanistic understanding: To understand the principles of visual processing, I develop and compare image-computable models that transform images into perceptual decisions.

Principled and interpretable models: A principled model is optimal with respect to a well-defined objective (cost function) under explicit, well-defined task-imposed and experimenter-imposed constraints, making its assumptions and thus its strengths and weaknesses rigorously specified. An interpretable model has clearly specified computational components whose roles in achieving the overall objective are transparent.

Scaling: Such models are typically available only for narrowly defined, highly constrained (simple) tasks. By scaling, I mean (i) developing this class of models for relatively complex tasks that incorporate properties of natural settings, and (ii) designing models that generalize across variations of the same task.

The advantages

Posing novel questions about mechanisms of visual processing (e.g. Oluk et al., 2023)
Allows us to explore the model space rigorously by thinking about the constraints and biologically plausible approximations (e.g. Oluk & Geisler, 2024)
Provides a benchmark to compare human and model performance, helpful for ML models where benchmarks are unavailable (e.g. Oluk & Geisler, 2025)
Applicable to a broad range of problems (e.g. 3D perception, motion perception, target identification)

Examples

1. Target Identification under high levels of uncertainty

Oluk, C. and Geisler, W. S. (2025). Evaluating ideal observers for large target identification tasks under additive white noise. 77th European Conference on Visual Perception, Mainz, Germany. Poster

Oluk, C., & Geisler, W. S. (2025). Target identification under high levels of amplitude, size, orientation and background uncertainty. Journal of Vision, 25(2), 3. GitHub Page.

Oluk, C., & Geisler, W. S. (2023). Effects of Target-Amplitude and Background-Contrast Uncertainty Predicted by a Normalized Template-Matching Observer. Journal of Vision, 23(12), 8. GitHub Page.

2. Perception of 3D orientation and depth of a planar surface

Oluk, C., Bonnen, K., Burge, J., Cormack, L. K., & Geisler, W. S. (2022). Stereo slant discrimination of planar 3D surfaces: Frontoparallel versus planar matching. Journal of Vision, 22(5), 6-6. GitHub Page.

Oluk, C., & Geisler, W. S. (2020). Ideal observers for the estimation of disparity in random-pixel stereograms. Journal of Vision, 20(11), 578-578. GitHub Page.

3. Perception of the global motion direction (ongoing)

Oluk, C., Szinte, M., Masson, G. S., Mamassian, P. (2024). Confidence responses in global motion discrimination task are well predicted by visual reliability. Vision Science Society Meeting, Florida, US. Poster

4. Perception direction and magnitude of the vernier presented in a motion stream (ongoing)

Holdsworth, M., Menétrey, M. Q., Oluk, C., and Herzog, M. H. (2025). A continuous adjustment task allows for better understanding of visual feature integration in a Sequential Metacontrast Paradigm. 77th European Conference on Visual Perception, Mainz, Germany.

2. Beyond Visual Perception

I am also broadly interested in human behavior and its underlying neural mechanisms. My approach can be easily extended to tasks where there are non-visual mechanisms (e.g. confidence, awareness) also leading the behavior in addition to perceptual mechanisms.

Examples

1. Confidence in perception of the global motion direction (ongoing)

2. Awareness of the perception of the vernier presented in a motion stream (ongoing)

3. Theory for model selection

The theoretical model space for each task is vast. With recent advances in machine learning and computational resources it is now far easier to develop mechanistic models, which makes the practical model space large. Thus, we rely on secondary principles such as simplicity, normative criteria, and assumptions about model type to identify models that are more likely to be human like. I am interested in formalizing these model selection and development procedures, emphasizing generalizability and scalability as prerequisites.

Examples

1. Efficiency as a model selection criteria (ongoing)