Subject: advice on postdoc (fwd) From: Al Bregman <bregman(at)HEBB.PSYCH.MCGILL.CA> Date: Wed, 30 Apr 1997 15:52:43 -0400
Dear list members, I just received this query from Dr. Schnitzler in Tuebingen, Germany. It concerns the question of where good signal-processing research on auditory neurobiology is being carried out where one of his recent graduates might go as a post-doc (with his own money). I'm not so familiar with this area; so I thought I'd post it to the list. You could answer directly to Dr. Schnitzler or, if you prefer, I could forward your replies to him. Thanks, Al ---------- Forwarded message ---------- Date: Wed, 30 Apr 1997 18:22:13 +0200 From: Hans-Ulrich Schnitzler <hans-ulrich.schnitzler(at)uni-tuebingen.de> To: bregman(at)hebb.psych.mcgill.ca Subject: advice on postdoc Dear Dr. Bregman please excuse the informal approach. When looking for a postdoc position for one of my students I contacted Mark Konishi for help. He recommend that I address you with the following question: Rolf Mueller, a PhD student in my lab is looking for a postdoc position in the states. In spring 1998 he will finish his dissertation in which he studies the possible role of acoustical flow field information in transfer flight and obstacle avoidance in CF bats. He is particularly interested in signal processing especially in simulating the function of sensory systems numerically. He is one of the best students I ever had and very gifted for this type of work. Additionally he works very hard and learns very fast. I am absolutely sure that it will be no problem for him to get a two year postdoc stipend for the states. We are looking for a top lab where he can improve his knowledge in signal processing and computer simulations. If possible he would prefer a lab which works on auditory information processing but he is also open for other fields (see the included personal profile of Rolf Mueller). Unfortunately I do not know the top labs working in the wanted fields. Do you have any suggestions where he could apply? Please find a short account on Rolf Mueller, his work and his interests, below. I would appreciate it very much if you could be so kind as to advise us on this issue. Sincerely, H.-U. Schnitzler -- Prof. Hans-Ulrich Schnitzler Dept. Animal Physiology, University of Tuebingen, Morgenstelle 28, D-72076 Tuebingen, Germany Phone +49 70 71 29 75 34 5; Fax +49 70 71 29 26 18 hans-ulrich.schnitzler(at)uni-tuebingen.de http://www.uni-tuebingen.de/tierphys/personen/schnitzl.htm Rolf Mueller - Personal profile Mini-CV I have studied biology at Tuebingen University since 1990; I specialized in neurobiology, supplemented by electronics and genetics. My Master- and PhD-thesis are detailed below. My thesis work is supported by a grant from Studienstiftung des Deutschen Volkes as were my undergraduate studies. I am also participating in the graduate program neurobiology at Tuebingen University. Master thesis (completed December 1995): Human listening experiments on the psychophysics of ranging paradigms. Scientific content (abstract): The ranging capabilities of the bats' sonar receiver have been studied extensively in behavioral 2-AFC - experiments employing the playback of artificial echoes to the animals' echolocation pulses. Such procedures are thought to generate the impression of phantom targets located at different distances. These tests made use of two paradigms: range difference (RD), the discrimination between different echo delays, and jitter versus non-jitter (JNJ), the detection of a regularly alternating, 'jittering' delay against a reference of constant delay values. The perceptional equivalence of these paradigms hypothesized by some workers has been doubted by others and a third paradigm, jitter-versus-jitter (JJ), which would require the discrimination of different jitter amplitudes, has been proposed for the purpose of clarifying this unresolved issue. In order to obtain a first glimpse at the psychophysical properties of the stimuli delivered in the ranging experiments, passive listening tasks with sequences of repeated artificial 'pulse-echo-pairs' where designed for human subjects according to all three mentioned rules and thresholds were measured in an adaptive 3-AFC procedure. It could be demonstrated, that the subjects' performance depended on a variety of stimulus parameters (e.g., repetition rate, reference value of the pulse-echo-delay and the characteristics of the individual pulses). The nature of these relationships depended on the tested paradigm, hinting at possible differences in the involved mechanisms. Each paradigm seems to constitute an individual perceptual task, with a somewhat closer relationship existing between JNJ and JJ. One major finding was, that the JNJ-performance depended on the pulse-echo-delay of the reference stimulus, which weakens the view, that a grossly similar effect found in bats can be regarded as proof for a specific internal representation of the stimuli along a range axis in these animals. In several experiments qualitative, individual differences among the subjects surfaced, which might be interpreted as a hint towards the existence of individual strategies. This would imply, that the solution of these tasks allows for different approaches. Experimenting with human subjects in order to elucidate the psychophysical properties of echolocation tasks, has it's limitations, however: The subjects of this study were carrying out passive hearing tasks, whereas ranging bats are making use of an active sensory system. Furthermore, modifications to the stimuli were necessary for adapting to the requirements of the human subjects. Therefore, no attempt is made to purport that problems associated with experiments in bats have been solved by this study once for all. Rather, these results should be regarded as hypotheses, which may be considered worth for further testing in bats. Methodological qualifications The experimental setup was entirely put together by me (on the basis of a preexisting function library for the DSP, the stimuli were generated on). It included: Implementation of the adaptive staircase procedure, the user interface, interfacing to a DSP-board and generating the stimuli there, as well as controlling the properties of the delivered stimuli. Furthermore, I did have a closer look at adaptive psychophysical methods and ran some Monte Carlo simulations of adaptive staircaises. Publications: Posters at: 10th International Bat Research Conference, Boston, 1995 and 24th Goettingen Neurobiology Conference, 1996 Paper: in prep. PhD thesis (to be completed spring 1998): The concept of the acoustical flow field and obstacle avoidance in CF-bats Scientific content (abstract): Bats are capable of obstacle avoidance based solely on echolocation information. The most likely candidate for the underlying perceptual process would in general be a combination of ranging by means of time-of-flight measurements and binaural direction-of-arrival determination. While it is beyond reasonable doubt, that these cues are involved in bats' space perception, some species may exploit other signal parameters conveying information about the spatial layout of their environment, which could be more readily evaluated given certain pulse designs and spatial tasks to be performed. In looking for alternative solutions to this problem, I choose the group of so-called CF-bats (CF = constant frequency) as a model system. These species can be observed to travel at high flight speeds (3-7 m/s) through obstacle rich forest habitats, maintaining an approximately constant height above ground. This situation may thus be thought of as a 2-D obstacle avoidance task. The signal design of the animals' echolocation pulses (narrow-band, hence the name, and comparatively long duration) does not lend itself easily as a substrate for range and angle estimation in the manner depicted above. The presence of multiple targets might be aggravating this difficulty by causing a resolution and a correspondence problem among the individual echoes received at each ear. I therefore explored the informational content of the proportional changes in center frequency (due to Doppler shifts) and amplitude (due to geometric attenuation, absorption and compound directivity of emitter and receiver) of the echoes in a rough, abstract-functional analogy to the time-to-contact variable of the optic flow field. It can be shown, that a combination of these two putative sensory variables allows in principal for a metrical reconstruction of the target position relative to the animals flight vector within a hemi-plane, i.e. symmetrical solutions to the right and to the left are found. A closer examination of this inverse problem reveals, that proportional frequency and amplitude changes by no means constitute an optimal foundation for a geometrical reconstruction of obstacle position: The cosine mapping angle to Doppler shift has minimal slope straight ahead, thus adversely affecting the precision of position estimation, where obstacle avoidance presumably requires most of it. Furthermore, depending on the assumed compound directivity function of receiver and emitter, positional ambiguities might be introduced. It should be noted, however, that optimality is not a necessary precondition of physiological plausibility and the error in positional judgment associated with a given precision of estimating the sensory variables' magnitude can be found to decrease as a function of distance to target, thereby accommodating the task's demands. Along the same line of thought, it may be argued, that the ambiguities eventually introduced by the directivity are confined to single positional estimates and could be pruned by looking at successive measurements. I arrived at the tentative conjecture, that an adequate obstacle avoidance maneuver, as observed in the animals, based on echolocation is most likely to require a measure for the situation's gravity (i.e. distance of obstacle to flight path) as well as urgency (i.e. distance to obstacle along flight path). The need for the latter may be attributed to the fact, that target strength could differ considerably among obstacles, thus affecting the extent of the echolocation system's field of view. Since detection distances are presumably rather short in general and eclipses during the inter-pulse-intervals restrict access to information about the target further, some representation equivalent to a knowledge of the target position relative to the flight path, seems highly desirable. Having demonstrated, that a metrical reconstruction is principally feasible (albeit not necessarily performed by the animals), there is good reason to believe, that both demands can be met by evaluation of our postulated sensory variables. In order to investigate, whether the information-bearing signal parameters are sufficiently well preserved in the auditory system's primal sketch, synthetic echo trains endowed with the proportional changes in frequency and amplitude pertinent to a given target position were generated and passed through a gamma tone filter bank adjusted to the neurophysiologically documented characteristics of CF-bat's. Then a reconstruction of target position is undertaken, with devoting special attention to the robustness against noise. Specifically, I evaluated how the auditory systems temporal integration might be able to subserve noise-reduction. The pending two parts of the project are to deal with multi-target situations and look at how the properties of my scene reconstruction hypothesis might relate to the requirements of interfacing with the execution of avoidance maneuvers. Methodological qualifications I have been employing a fair range of numerical methods (solving linear and non-linear systems of equations, numerical integration, optimization, etc.) during my work. I tried to acquaint myself with the theory of signal processing, i.e. linear system theory (laplace and z-transform), digital filtering and spectral estimation. Implementations were done mostly using matlab (symbolic maths was carried out with maple), but some stand-alone c-code has been written also. All work was done on a unix-platform (linux for i386). Publications: Poster at: 25th Goettingen Neurobiology Conference, 1997 Presentation accepted for 33rd ASA Meeting, State College, Pennsylvania Postdoc profile: The research I would like to participate in should address signal processing algorithms as models for the function of the brain's sensory systems. It could involve mimicking these functions in some artificial system (numerical simulation or maybe even a physical realization). The coupling of perception and action falls within the scope of my interests, as well. I have a keen interest in improving my theoretical background knowledge. I would like to run computer simulations, psychophysical experiments would also interest me. I would be comfortable doing computer simulations without psychophysics on human observers, but not vice-versa, since it is my aim to have my research on sensory systems' function resting on a profound theoretical background. Audition as a model sensory modality would be preferred, but I could imagine participating in any research fitting under the above specifications, since I would like to be working more on the side of abstract function principles. Biomimetic sonar is still on my wish list, but not of any priority.