We are interested in understanding the mechanisms underlying the primate visual system--a system to which nearly half of our cerebral cortex is devoted. The massive complexity of this system is belied by the apparent ease with which we can:

  • recognize the face of a friend
  • drive at high speeds on a crowded highway
  • visualize the particular shade of blue in our carpet as we look for a matching color of paint in the store
  • identify someone's handwriting
  • track an oncoming baseball
There are many appraches, both experimental and theoretical, that are currently used to study the visual system; most of these approaches are used in our lab or in our collaborators' labs. Together, the various approaches outline the steps of the scientific method-- 1) psychophysics and neuropsychology suggest constraints on the behavior of a visual subsystem, 2) electrophysiology and functional brain imaging give constraints on the architecture of the subsystem, 3) together these suggest a particular computer model, and finally 4) issues confronted during the implementation of the computer model may suggest new directions for experimental research. Following is a brief description of each of these techniques, along with a summary of ongoing work in each area.

Psychophysics. Psychophysics uses very detailed measurements about people's ability to perform a visual task in order to infer something about the brain mechanism responsible for that task. This often involves measuring a threshold--for example, what is the smallest detectable change in brightness of a light source? By identifying the limits of a system, one learns more about how the system operates in normal ranges. A main focus of our psychophysical investigations is selective visual (focal) attention in normal subjects, using techniques from sensory psychology, in particular the dual-task paradigm (Dr. Achim Braun, Dr. Geraint Rees, & Dr. Barbara Zenger). In addition, Rob Peters is starting a project on visual object recognition in humans, carried out in tandem with Dr. Fabrizio Gabbiani and Prof. Nikos Logothetis who are investigating the exact same object recognition paradigm in monkeys.

Electrophysiology. This technique involves making electrical recordings from individual neurons by inserting a metal or glass electrode directly into the brain. For obvious reasons, this technique cannot typically be used with humans, so most electrophysiological data comes from monkeys. However, some limited electrophysiological data are becoming available from human patients with epilepsy, who often have electrodes temporarily inserted into their brains as part of their medical treatment.Gabriel Kreiman is collaborating with Dr. Itzhak Fried (UCLA) to use such human electrophysiological data to study the roles of the temporal lobe and nearby subcortical structures in object recognition. We have recorded from several multiple units in the human medial temporal lobe. We characterized the visual responses showing that single neurons respond selectively to complex stimuli including faces, objects and spatial layouts. We also recorded the neuronal activity while the subjects had to imagine the stimuli with the eyes closed. We found that single neurons changed their activity selectively depending on the stimuli they were recreating on their "mind's eye". Furthermore, most of these neurons had the same selectivity during imagery and vision. See Kreiman, G., Koch, C. & Fried, I. Category-specific visual responses of single neurons in the human medial temporal lobe. Nat. Neurosci. 3, 946-953 (2000). Kreiman, G., Koch, C. & Fried, I. Imagery neurons in the human brain. Nature (In Press)

Neuropsychology. This discipline attempts to make associations between brain regions and specific functionality by studying the symptoms of patients with damage to their visual system due to stroke or other injury or disease. It is important to discover not only which visual abilities are lost when a region is damaged, but also which abilities are retained.

Functional brain imaging. The last several years have seen an explosion of research in the area of functional brain imaging, typically function magnetic resonance imaging (fMRI). This technique has several advantages over electrophysiology: it is a non-invasive procedure that can be done with normal human subjects, and it allows the entire brain to be studied at once, rather than one neuron at a time. However, it has its disadvantages as well: it has poor spatial and temporal resolution compared to electrical recordings, and it does not directly measure electrical activity in neurons, but rather a side effect (blood oxygenation level). In collaboration with Linda Chang, M.D. and Thomas Ernst, Ph.D. at UCLA-Harbor Medical Center, we are using fMRI to investigate the neuronal basis of object recognition in human subjects (Dr. Jorge Jovicich, a joint post-doctoral fellow with Drs. Chang and Ernst; Rob Peters). Dr. Geraint Rees continues to use fMRI to investigate the neural bases of selective attention and visual awareness in collaboration with the Functional Imaging Laboratory at University College London.

Computational modeling. Understanding complex information-processing tasks, such as the spatial perception, motion or selective, visual attention, requires a firm grasp of how the problems can be solved at the computational level, and how the resulting algorithms can be implemented onto the known architecture of the striate and extrastriate cortical areas (as well as associated subcortical areas) in the primate visual system. By tweaking the parameters of computer models to compare their behavior with data from psychophysics, electrophysiology, or functional brain imaging, we gain an understanding of the important variables in biological vision.

We use analytical methods, coupled with detailed neural network simulations of the appropriate circuitry, to model many of those same visual subsystems that we study with experimental approaches: motion perception, selective visual (focal) attention (Dr. Steffen Egner, Laurent Itti, Dr. Geraint Rees, Dr. Barbara Zenger, Dr. Jochen Braun), object recognition (Dr. Fabrizio Gabbiani, Rob Peters). For a cool Java applet demo of our saliency-based attention-system, click here.

Visual illusions. Our laboratory exploits visual illusions, such as the Breathing Square Illusion (Java applet demo), as a window onto the algorithms and neural mechanisms underlying visual perception in primates. Al Seckel is editing and researching an extensive collection of such illusions (many of them non-visual), as well as illusory artforms and making them available to the general community via a beautiful and highly interactive website.

[Home] [News] [Research] [People] [Papers] [Classes] [Jobs]