Past events

In this talk I will show how complex visual tasks can be performed by exploiting redundancy in visual data. Comparing and integrating data recurrences allows to make inferences about complex scenes, without any prior examples or prior training. I will demonstrate the power of this approach to several visual inference problems (as time permits). These include: 1. Detecting complex objects and actions (often based only on a rough hand-sketch of what we are looking for). 2. Summarizing visual data (images and video). 3. Super-resolution (from a single image). 4. Prediction of missing visual information. 5. Detecting the "irregular" and "unexpected". 6. "Segmentation by Composition".

The eating disorders anorexia nervosa (AN) and bulimia nervosa (BN) are serious psychiatric disorders characterized by disturbed eating behavior and characteristic psychopathology, and in the case of AN, very low weight. The mortality of AN is the highest of any psychiatric disorder. The etiology of AN and BN are multidetermined; there are factors biologically, psychologically and socioculturallly that predispose an individual to an eating disorder. Biologically, genes contribute significantly to the risk for eating disorders. Studies have shown that the risk of anorexia nervosa in first degree relatives if one parent has AN is between 8-10%.compraed to the general population risk of 1%. The concordance rate for MZ twins in AN is close to 70%. Approximately 70% of the variance in AN is attributable to genetic effects whereas about 30% is attributable to unique environmental effects. For BN, approximately 60% of the variance in BN is attributable to genetic effects whereas about 40% is attributable to unique environmental effects. Eating disorders do not map on to one chromosome Instead there are dimensions that are genetic, such as risk of obesity, anxiety, and temperament such as perfectionism and obssessionality that are inherited and place an individual at risk for an eating disorder Gender is also a genetic risk factor for an eating disorder. Being female is a risk factor for an eating disorder, not just because females are more sensitive to cultural pressure than males. Females are more commonly affected by eating disorders because female brains are much more sensitive to dietary manipulation than male brains related to the effects of estrogen and progesterone on serotonin metabolism. Tryptophan depletion does not significantly affect levels of brain serotonin in males but dramatically reduces levels of serotonin produced in females’ brains. Dieting, especially restricting carbohydrates lowers the level of blood tryptophan available to cross the blood brain and be available to be synthesized to serotonin. Patients are at risk for an eating disorder will reduce the levels of serotonin produced in their brains by dieting and restricting carbohydrates, leading to changes in satiety and mood and increasing the likelihood of an eating disorder developing . There are those who believe that binge eating develops in response to a hyposerotonergic state in an attempt to restore tryptophan available for brain serotonin synthesis I have been involved in several large multi site genetic studies of eating disorders over the past 15 years. In a linkage analysis of affected relative AN pairs, when only restricting anorexics were included in the analyses, a significant signal was found on the long arm of chromosome 1. Candidate genes that have been found in that area of chromosome 1 include the serotonin 1D receptor gene, the opiod delta gene, and the dopamine D2 receptor gene. In a linkage analyses on a sample of affected relative pairs with BN, a significant signal was found on chromosome 10 when the sample included only subjects who vomited. I am currently involved in a whole genome wide association study ( GWAS) of 4000 AN cases and 4000 female controls which will hopefully elucidate which specific genes contribute to the risk for AN. Future genetic studies we are involved in will focus on why patients with AN are able to drop their weight to dangerously low levels, whereas patients with bulimia nervosa (BN) with similar psychopathology and dysfunctional eating behaviors are protected from extreme weight loss and do not develop AN. So far, research on genes that are important for appetite and weight regulation, such as the leptin receptor (LEPR), ghrelin (GHRL), melanocortin 4 receptor (MC4R), and brain derived neurotrophic factor (BDNF), has yielded conflicting findings in AN and BN, while related genes with potential in the same genetic systems have not been sufficiently studied. Considering that AN in adults tends to follow a chronic course and currently does not have any evidence-based treatments, determining the role of genetic factors in the vulnerability to achieve low weight in AN patients could be an important first step toward improved treatment.

Our visual system is capable of inferring the structure of 2-d images at a resolution comparable (or, in some tasks, greatly exceeding) the receptive field size of individual retinal ganglion cells (RGCs). Our capability to do so becomes all the more surprising once we consider that, while performing such tasks, the image projected on the retina is in constant jitter due to eye and head motion. For example, the motion between two subsequent discharges of a foveal RGC typically exceeds the receptive field size, so the two subsequent spikes report on different regions of the visual scene. This suggests that, to achieve high-acuity perception, the brain must take the image jitter into account. I will discuss two theoretical investigations of this theme. I will first ask how the visual system might infer the structure of images drawn from a large, relatively unconstrained ensemble. Due to the combinatorially large number of possible images, it is impossible for the brain to act as an ideal observer that performs optimal Bayesian inference based on the retinal spikes. However, I will propose an approximate scheme derived from such an approach, which is based on a factorial representation of the multi-dimensional probability distribution, similar to a mean-field approximation. The decoding scheme that emerges from this approximation suggests a neural implementation that involves two neural populations, one that represents an estimate for the position of the eye, and another that represents an estimate of the stabilized image. I will discuss the performance of this decoding strategy under simplified assumptions on retinal coding. I will also compare it to other schemes, and discuss possible implications for neural visual processing in the foveal region. In the second part of the talk I will focus on the Vernier task, in which human subjects achieve hyper-acuity, greatly exceeding the receptive field size of a single RGC. The optimal decoder for this task can be formalized and analyzed mathematically in detail. I will show that a linear, perceptron-type decoder cannot achieve hyper-acuity. On the other hand a quadratic decoder, which is sensitive to coincident spiking in pairs of neurons, constitutes an effective and structurally simple solution to the problem. Furthermore, the performance achieved by such a decoder is close to the limit imposed by the ideal Bayesian decoder. Therefore, spike coincidence detectors in the early visual system may facilitate hyper-acuity vision in the presence of fixational eye-motion.

Formation of neural circuits requires that axons recognize appropriate cells, and even appropriate parts of cells, upon which to synapse. In the retina, amacrine and bipolar cells form synapses on retinal ganglion cells (RGCs) in the inner plexiform layer (IPL). The visual features to which different RGC subtypes respond depend on what input they receive, prime determinants of which are the IPL sublaminae in which their dendrites make synapses. We have therefore sought molecules that mark RGC subtyoes and mediate lamina-specific connectivity. Candidates include members of the immunoglobulin superfamily, such as Sidekicks, Dscams and JAMs, and members of the cadherin superfamily, such as Class II and protocadherins. I will discuss our progress toward identifying and testing such candidates. I will also discuss methods for tracing connections of retinal neurons in wild-type and mutant mice, so that we can assess the consequences of perturbing target recognition systems.

Fear is a crucial adaptive component of the behavioral repertoire that is generated in relation to stimuli which threaten to perturb homeostasis. Fear-relevant associations are learned and consolidated as part of long term memory. After learning, fear responses are modulated through processes termed safety learning and extinction. Perturbation of these mechanisms can lead to disproportional anxiety states and anxiety disorders. Recent years have seen considerable progress in identifying relevant brain areas – such as the amygdala, the hippocampus and the prefrontal cortex - and neurophysiological principles. Key mechanisms, involving rhythmic oscillations of neuronal subpopulations and neuromodulatory influences, will be discussed

Understanding the neural mechanisms of visual object recognition is a difficult task in part, because for any given object it is not clear, which exact part of the brain to study. Yet evolution has presented us with a unique model system to decipher these mechanisms. The temporal lobes of macaque monkeys contain neural machinery to support face recognition consisting of six discrete patches of face-selective cortex. The two main organizing features of this system – concentration of cells encoding the same complex object category into modules and spatial separation of modules – make it possible to break down the process of face recognition into its components. In my talk I will present anatomical results supporting the notion that the distributed face patches really are part of an integrated face-processing machine, and electrophysiological results showing that each patch subserves a distinct computational function. In the second part of my talk I will turn to something completely different, attention. Using fMRI in macaque monkeys, we found a network of areas to be modulated by attention in motion-discrimination task, included a hitherto unsuspected region within inferotemporal cortex, PITd. We then targeted PITd for electrophysiological recordings and electrical microstimulation in different tasks to learn about its role in sensory information processing and spatial attention. I will discuss the somewhat radical conclusion we arrived at, namely that PITd may constitute a region for attentional control.