MARK CHANGIZI

Principles governing the organization of city road networks...and the brain
Cities and the mammalian neocortex may seem to have little in common, but each is approximately a surface with a network of conduits (roads and neurons, respectively) connecting its disparate parts. Because both cities and brains are under selection pressures to make their connections efficiently, we investigate the hypothesis that the organization of city highway networks and the mammalian neocortex may be governed by common principles. Here we measure how city highway networks vary with city size and find that, consistent with the hypothesis, highway networks scale with exponents nearly identical to those found for the analogous quantities in the neocortex. As a function of surface area, the number of conduits scales approximately as the 3/4 power, the number of "leaves" (highway exits and synapses) scales approximately as the 9/8 power, propagation velocity scales approximately as the 1/8 power, and total conduit surface area scales approximately as the 11/8 power. We also find that city population scales as the 1.46 power of surface area, potentially driven by the total surface area of highways. We discuss the extent to which explanations for neocortical scaling can be extended to cities. Press has included: LiveScience (Yahoo News), Daily Planet TV show (about half way), ScientificBlogging, Reason, The Atlantic, Open Mag, io9, Digital City, Tyden, Wissenschaft, Donbass, Tendencias, RPI

Perception

Techniques for early visual detection of oxygen desaturation
Central cyanosis refers to a bluish discoloration of the skin, lips, tongue, nails, and mucous membranes, and is due to poor arterial oxygenation. Although skin color is one of its characteristic properties, it has long been realized that by the time skin color signs become visible, oxygen saturation is dangerously low. Here we investigate the visibility of cyanosis in light of recent discoveries on what color vision evolved for in primates. We elucidate why low arterial oxygenation is visible at all, why it is perceived as blue, and why it can be so difficult to perceive. With a better understanding of the relationship between color vision and blood physiology, we suggest two simple techniques for greatly enhancing the clinician's ability to detect cyanosis. Here is a piece in Forbes, and a piece I wrote in ScientificBlogging.

Complexity

Medical Hypotheses

A new function for binocular vision, and the evolution of forward-facing eyes
Why do our eyes face forward, and why do many animals have eyes facing sideways? Here we describe new research suggesting that the degree of binocular convergence is selected to maximize how much the animal can see in its environment. Animals in non-cluttery environments can see the most around them with panoramic, laterally directed eyes. Animals in cluttery environments, however, can see best when their eyes face forward, for binocularity has the power of "seeing through" clutter out in the world. Evidence across mammals closely fits the predictions of this "x-ray" hypothesis, and is hard to reconcile with traditional explanations where stereo vision is critical. News stories include the following: Le Point, Science Daily, Caltech Press Release, Times of India, Wissenschaft, Epoch Times, Spiegel, Membrana, Polska, Real Science (audio). The idea was also touched upon in my Scientific American interview, and in Rensselaer Magazine.

Journal of Theoretical Biology

The underlying economic rationality of preference and affect dynamics
Visual exposure to an object can modulate an observer's degree of preference for it, initially enhancing preference (a "familiarity preference" regime), and eventually lowering it again (a "novelty preference" regime). Here we investigate whether there may be a functional advantage to modulating preference in this way. We put forth the simple hypothesis that degree of preference for an object of type X is the brain's estimate of the expected value of acting to obtain X. In light of this view of what preferences fundamentally represent, we are able to explain the "exposure effect" and many of the connected phenomena. Some press: Live Science, Fox News, Scientific Blogging, Forskning, Atelier, Express. RPI.

Journal of Theoretical Biology

The Science of Social Vision. New York

Perception

Bare skin, blood, emotion, and the evolution of primate color vision
The primate face undergoes color modulations (such as blushing or blanching), some which may be selected for signaling and some which may be an inevitable consequence of underlying physiological modulations. Because for highly social animals like most primates, one of the most important kinds of object to be competent at perceiving and discriminating is other members of one's own species, we hypothesized that primate color vision has been selected for discriminating the short term spectral modulations on the skin of conspecifics, these modulations providing useful information about the current state or mood of another conspecific. Here we show that for the two dimensions of skin spectral variation in the short term, the dimension due to the fraction of blood in the skin corresponds approximately to the blue-yellow opponent channel (more blood ==> bluer), and the other dimension due to oxygen saturation of the blood corresponds approximately to the red-green opponent channel (greater oxygenation ==> redder). Trichromats, but not dichromats, are therefore sensitive to both dimensions of skin color variation, and, more specifically, the wavelength sensitivities of the M and L cones for trichromatic primates are near-optimal for sensing modulations of oxygen saturation. Also, because skin color modulation cannot be seen on a furry face, trichromatic primates tend to have bare faces. See Reuters, New Scientist, Financial Times, Scientific American, Scientific American mention, Time Magazine, Discover Magazine, Rensselaer Magazine, Bild der Wissenschaft (roughly a German Scientific American), Daily Telegraph, Daily Telegraph (Santa), American Scientist, BoingBoing, ABC News, Pasadena Star News, BBC Wildlife Magazine, Bluesci (Cambridge Science Magazine), Ingenioren, Der Standard, The Times of London, Rhein Zeitung, Kagaku (Japanese science magazine), Nikkei Science, CBC News, The Independent (London), Best Friends Magazine, Die Presse, Ego-Net, GEO Magazine, 3SAT, Die Welt, Iran Daily, 123, Arkadas, Ceske Novinky, Asahi Shimbun, CNet, 24 ThoiSu, Impact, Fugle og Natur, Anthropology.net, Complexity Digest, Softpedia News, Erkenntnisse des Neuromarketing, The Telegraph (Calcutta, India), Nana 10, Nana (navel) (English translation), and Caltech News for news stories about this (some which were picked up and translated worldwide, e.g., the New Scientist, Reuters, Pasadena Star, and Caltech News stories). Also see CBC "As It Happens" for a radio interview (go 22 minutes into RealAudio clip), and here is the script of a public radio show called Loh Down on Science that covered this research (and here is the RealAudio clip). See also Science Magazine for some controversy.

The Science of Social Vision. New York

A general theory of mammalian visual cortex organization
The part of the primate visual cortex responsible for the recognition of objects is parcelled into about a dozen areas organized somewhat hierarchically (the region is called the ventral stream). Why are there approximately this many hierarchical levels for object recognition? Here I put forth a generic information-processing hierarchical model for visual object recognition, and show how the total number of neurons required depends on the number of hierarchical levels and the complexity of visual objects that must be recognized. Because the recognition of written words appears to occur in a similar part of inferotemporal cortex as other visual objects, the complexity of written words may be similar to that of other visual objects; for this reason, I measure the complexity of written words, and use it as an approximate estimate of the complexity of visual objects more generally. I then show that the information-processing hierarchy that accommodates visual objects of that complexity possesses the minimum number of neurons when the number of hierarchical levels is approximately 15 and when the sizes of areas decrease exponentially with level, each level on average approximately 1.25 times larger than the level above it. I show that these optimal properties are close to those found in the primate ventral stream.

Biology Letters

Biological Cybernetics

The economical organization of the lexicon
Good definitions consist of words that are more basic than the defined word. There are, however, many ways of satisfying this desideratum. For example, at one extreme, there could be a small set of atomic words that are used to define all other words; i.e., there would be just two hierarchical levels. Alternatively, there could be very many hierarchical levels, where a small set of atomic words is used to define a larger set of words, and these are, in turn, used to define the next hierarchically higher set of words, and so on to the top level of very specific, complex words. Importantly, some possible organizations are more economical than others in the amount of space required to record all the definitions. Here I ask, How economical are dictionaries? Here I present a simple model for an optimal set of definitions, predicting on the order of 7 hierarchical levels. I test the model via measurements from WordNet and the Oxford English Dictionary, and find that the organization of each possesses the signature features expected for an economical dictionary. See a longer blurb of this here. Some stories in the press (many which were picked up all over the world): Scientific American, Tendencias (Spain), RPI News, Red Orbit, New Kerala, United Press International.

Journal of Cognitive Systems Research

Why letters are shaped the way they are
Are there empirical regularities in the shapes of letters and other human visual signs, and if so, what are the selection pressures underlying these regularities? To examine this, we determined a wide variety of topologically distinct contour configurations, and examined the relative frequency of these configuration types across non-logographic writing systems, Chinese writing, and non-linguistic symbols. Our first, and main, result is that these three classes of human visual sign possess a similar signature in their configuration distribution, suggesting that there are underlying principles governing the shapes of human visual signs. Second, we provide evidence that the shapes of visual signs are selected to be easily seen, at the expense of the motor system. Finally, we provide evidence to support an ecological hypothesis that visual signs have been culturally selected to match the kinds of conglomerations of contours found in natural scenes, because that is what we have evolved to be good at visually processing. There are press stores about it at the Sciam Mind, The Atlantic, Sciam, Barnes and Noble review, NY Times (review of Dehaene's book, discussing my research as "most interesting"), WSJ (discussed in review of Dehaene's book), Daily Telegraph, Rensselaer Magazine, USA Today, Newsweek (print and online), NRC Handelsblad, Cahiers de Science et Vie, Australian Broadcasting Company, Mokslo Lietuva, Live Science, Suddeutsche Zeitung, Columbia Tribune, De Morgen, Softpedia News, Svoboda News, USA Today Tech Space, Netinfo.bg Bulgaria, Internet Haber, Nikolaev, Engineering and Science Magazine, Caltech News and EurekAlert. Some of these news stories were picked up worldwide, such as Live Science (e.g., picked up by Fox News and MSNBC), Caltech News and EurekAlert. See CURJ for an article by undergraduate Qiong Zhang in the Caltech undergraduate research magazine.

How bigger brains are made
This research answers why the neocortex is folded, why the number of synapses per neuron increases, why white matter scales up disproportionately quickly, why the number of cortical areas increases, why soma and axon radius increase, and more. The theory explaining these features posits that the neocortex is well-connected in a wire-optimal fashion. The paper also estimates the network diameter of the neocortex to be about 2. Newer research appears in Section 1.1 of my book. Appearing soon is a contributed chapter (see below) in Evolution of Nervous Systems. Also, a new paper concentrating on scaling of parcellation and area-area connectivity has appeared in BBE.

The American Naturalist

Featured Article

New Encyclopedia of Neuroscience.

Evolution of Nervous Systems.

Brain, Behavior and Evolution

Biol Cybern

Patterns across writing systems, and what it tells us about visual recognition
A writing system is a visual notation system wherein a repertoire of marks, or strokes, is used to build a repertoire of characters. Are there any commonalities across writing systems concerning the rules governing how strokes combine into characters? In an effort to answer this question we examined how strokes combine to make characters in more than 100 writing systems over human history, ranging from about 10 to 200 characters, and including numerals, abjads, abugidas, alphabets and syllabaries from five major taxa---Ancient Near-Eastern, European, Middle Eastern, South Asian, Southeast Asian---as well as invented writing systems. We discovered underlying similarities in two fundamental respects. (1) The number of strokes per character is approximately three, independent of the number of characters in the writing system; numeral systems are the exception, having on average only two strokes per character. (2) Characters are approximately 50% redundant, independent of writing system size; intuitively, this means that a character's identity can be determined even when half its strokes are removed. Because writing systems are under selective pressure to have characters that are easy for the visual system to recognize and for the motor system to write, these fundamental commonalities may be a fingerprint of mechanisms underlying the visuo-motor system. See New Scientist, Natural History Magazine, Spiegel, Jay Ingram (from Discovery Channel), Wissenschaft, Wissenschaft-Online, ORF, Arzte Zeitung, San Diego Union-Tribune, GEO, Net Hirlap, and Asahi Shimbun for some of the news stories about our paper.

Proc Roy Soc Lond B

Latency correction and a general theory of illusions
Perceiving-the-present is the theoretical framework positing that the function of the visual system is to generate percepts representative not of the scene that generated the proximal stimulus, but of the scene that will be present at the time the percept actually occurs about 100 msec later, thereby compensating for the neural delay. To achieve this, the visual system must utilize ecological regularities to "guess" what is about to happen in the next moment. One of the most common kinds of ecological regularity is forward movement, and this research demonstrates that the visual system responds with appropriate latency-correction mechanisms when cues suggest forward movement. On the basis of this it is possible to predict a pattern of illusions over two dozen classes of stimuli. See this poster for a very brief look at the table unifying and explaining more than 50 kinds of illusion. In earlier work I showed how this idea explains the classical geometrical illusions, in particular. See also Chapter 2 of my book. See a popular article in Gehirn & Geist that touches on this, as well as press concerning the unifying theory as follows: Welt, ABC News, The New York Times, Spiegel, Scientific American, Rensselaer Magazine, 3sat Nano TV (German television show) [version w/out voice-over], 3sat (text for Nano television piece), FOX News Channel (live television interview), Albany's Channel 10 News (television interview), Live Science (picked up in Yahoo News, MSNBC, Fox News, and worldwide), Scientific American Mind, Newsweek(ru), Sync (Dutch), BoingBoing, Lufthansa Exclusive, RPI News, Caltech News, Newsland (Russian), Science Daily, Times of India, MSN India, Technocrat, Tarakosh Josh!, Lawrence Journal-World, TechRevu, DVICE, Technovelgy, Tennis Korea (reprint from ChoSun), INews24, Pasadena Weekly, Reporter Online, MSN (de) .

Behavioral and Brain Sciences

Problems of Space and Time in Perception and Action.

Cognitive Science

Perception

A general theoretical framework for complex, behavioral networks
In this research direction, I demonstrate that---and explain why---behavioral networks of all kinds tend to follow similar principles for (1) how nodes locally combine to implement network structures, (2) how structures combine globally to implement network-level behaviors, (3) how connectivity increases in larger networks, and (4) how larger networks become increasingly parcellated into distinct subregions. In the earlier papers, I consider the manner in which languages (e.g., English over the last 800 years, and child development of phonemes, words and sentences), bird song, and other such systems increase in complexity. In no case do there appear to be universal languages, in the sense that a single set of component, or word, types suffices for the construction of arbitrarily many expressions, or sentences. Also, I provide evidence that although bird song seems combinatorial, it is not. Finally, I show that, even though human language grammar does not constrain the length of sentences, there appear to be combinatorial limits to actual sentences, and this drives the vocabulary growth rate of the English language over time. See also Chapter 1 of my book. Later papers demonstrate that---and explain why---the relationship between number of node types and network size is a power law for many kinds of network, including networks of cells, neurons, ants, employees, and even Legos. We show that networks do not use universal languages (i.e., a fixed number of node types from which all network complexity may be achieved), but, instead, networks have invariant combinatorial degrees (i.e., the degree of combinatorialness allowed in building functional expressions from nodes). The scaling features reveal that human-constructed networks have low combinatorial degrees (on the order of two), whereas natural networks have high combinatorial degrees (from 5 to 15, depending on the kind of network). We also apply the theoretical ideas to ecosystems, providing for the first time a connection between food web features (food chain length) and species-area plot scaling exponents: archipelagos with food chain length L are expected to have scaling exponents of roughly 1/L. See also applications of these ideas to hierarchical evolution, with Daniel McShea. The most recent paper, in Complexity, provides a unifying framework for understanding four "correlates of behavioral networks" (this paper was listed in Complexity Digest).

Perception

Complexity

Integr Compar Biol

J Theor Biol

Comments
Theor Biol

J Theor Biol

Behavioral complexity in mammals
In this research I reveal the relationships between behavioral repertoire size, encephalization, and number of muscles in mammals. I demonstrate that muscles are, indeed, used in a combinatorial manner to implement behaviors---something not following from the mere fact that behaviors are built from multiple muscles. The research also reveals that behavioral repertoire size correlates well with encephalization, something nearly everyone believes, but, to my knowledge, no one has measured among mammals. A compilation of behavioral repertoire sizes for 28 species across six non-mammalian classes may be found in Chapter 1 of my book. The paper also shows, via new data, how behavioral complexity increases during the ontogeny of rat.

J Theor Biol

Why animals (and other biological entities) have as many limbs as they do
This research shows that there is a particular quantitative relationship between limb number and body-to-limb proportion across many animal phyla (for animals with radially, not ventrally, projected limbs). Namely, animals with long limbs relative to their body size tend to have on the order of six limbs, and as an animal's limbs become shorter relative to body size, the number of limbs increases in a particular quantitative fashion. I explain the relationship in terms of an optimality hypothesis. The idea is fleshed out in Chapter 1 of my book, and raw data are provided. Check out the interactive animal limb demo, which helps to visualize the relationship between number of limbs and body-to-limb proportion. See some discussion at Science.ca. See Tubitak Bilim ve Teknik for a press story on it. A second paper, in progress, demonstrates an inverse-square law for spherical biological structures with limbs, such as viruses and pollen.

In progress.

The shapes of neurons and arteries are volume-optimal
We demonstrate that neuronal and arterial trees (as well as some other kinds) have volume-optimal geometries. We treat source (e.g., soma) and leaf (e.g., synapse) locations as fixed, and ask, of all possible tree shapes reaching from the source to the leaves, which utilizes the least volume, and how do natural arbors compare to these volume-optimal ones? We demonstrate that a variety of neuronal trees and coronary arterial trees have these volume-optimal geometries. Furthermore, we argue that it is obtained via a simple self-organizing, fluid-mechanical mechanisms. We also show for the first time that neuron branch diameters conform to Murray's Law, where the cube of the trunk segment diameter is equal to the sum of the cubes of the diameters of the children.

Biol Cybern

Can J Physiol Pharmacol

Phys. Rev. E

Decision theoretic approaches to perception
Does thirst make you more likely to think you see water? Tales of thirsty desert travelers and oasis mirages are consistent with our intuitions that appetitive state can influence what we see in the world. Yet, there has been surprisingly little scrutiny of this appetitive modulation of perception. We tested whether dehydrated subjects would be biased towards perceptions of transparency, a common property of water. We found that thirsty subjects have a greater tendency to perceive transparency in ambiguous stimuli, revealing an ecologically appropriate modulation of the visual system by a basic appetitive motive. This also provides support for the Bayesian approach to visual perception because, taking the Bayesian approach seriously, there ought to be perceptual consequences to modifying the utilities without modifying the probabilities, something we show in the paper by varying the observer's appetitive state. See Trends in Cognitive Sciences and The Psychologist for two news stories about our paper, and an oblique reference in Science Magazine.

Learning hunger and thirst
Experiments with rats were conducted demonstrating that hunger and thirst behavior (i.e., the appetitive food and water seeking behaviors) are learned: Without food experience when food-deprived or water experience when water-deprived, deprived rats do not seek food or water, respectively, despite having experience with food and water when not deprived.

Perception

Riddle of induction
I give a kind of solution to the riddle of induction. Concerning a candidate set of hypotheses from which we would like to learn the probable one, and for which we currently have absolutely no data, if you tell me what properties you deem these hypotheses to have (this set of properties comprises your "conceptual framework"), then my theory will tell you what prior probabilities you ought to place on the hypotheses in that set (based on several symmetry-related principles). For a more recent presentation of the ideas, as well as applications to innateness in the brain, see Chapter 3 of my book.

Physiol Behav

Synthese

Vagueness of natural language
Here I provide an explanation for why natural language words are vague. I show that vagueness is something with which any finite, computational language-user must contend. (Vagueness is directly related to the sorites paradox, or the paradox of the heap. Premise 1: One grain of sand cannot make a heap. Premise 2: Adding one grain of sand can never turn a non-heap into a heap. Conclusion: There are no heaps. Clearly a false conclusion, and thus the paradox. Because the heap paradox relies on the vagueness of 'heap' to make it paradoxical, this research also serves as a resolution of this ancient paradox.) For my most recent work on the idea, see Chapter 4 of my book. The conference paper below differs from my other vagueness work in that it presents mathematical notions concerned with vagueness, such as that there are multiple degrees to which the borderline region of a vague word can be computationally ineliminable.

Acta Analytica

Mathematical inevitability of the eureka, or Aha!, phenomenon
There are some problems for which it is possible to know how close one is to solving it. I show that there is a tower of such "monitorable" problems, each requiring more resources to monitor. And then I show that there are computable problems that are not monitorable at all; the only way to solve these is via having a "eureka" moment. In addition to the published math paper below, I have an old unpublished non-mathematical manuscript that can be linked here.

Synthese

Ultimate computational limits on learning
Natural notions of "learning with error" are introduced, and the ultimate computational limits of learning studied in light of the new notions.

w^w

Information and Computation