Krishnan Kumaran

Visual organization for figure/ground separation

A common factor in all illusory contour figures is the perception of a surface occluding part of a background. In our previous work, we have shown we could diffuse a proper set of junction hypothesis (what is salient or background) to obtain a surface where their boundaries represented illusory contours. Amodal completions emerge at the overlapping surfaces. We address the problem of selecting the best image organization (set of hypothesis). We propose an optimization criteria based on a coherence measure between pairs of junctions (correlation between the diffusion of each pair). A statistical physics approach to select the best organization is applied. The experiments suggest that despite the large number of possible organizations our approach may take only a few steps (in organization space) to select the best one.

Visual Organization of Illusory Surfaces

A common factor in all illusory contour figures is the perception of a surface occluding part of a background. These surfaces are not constrained to be at constant depth and they can cross other surfaces. We address the problem of how the image organizations that yield illusory contours arise. Our approach is to iteratively find the most salient surface by (i) detecting occlusions; (ii) assigning salient-surface-states, a set of hypothesis of the local salient surface configuration; (iii, applying a Bayesian model to diffuse these salient-surface-states; and (iv) efficiently selecting the best image organization (set of hypothesis) based on the resulting diffused surface.

Simulation of self-organizing spectrum management in wireless networks

We describe the simulation of a new dynamic channel assignment algorithm in FDMA/TDMA wireless networks. The algorithm relies on periodic interference measurements by each of the base stations on the inactive frequencies, so as to identify appropriate candidate channels. The adaptive nature provides automatic configuration at the time of system initialization and adaptation to system expansion and traffic patterns with spatial or temporal variations. By eliminating the manual frequency planning process inherent to todays fixed channel assignment procedures, the self-organizing capability guarantees ease of operation for service providers, while increasing both capacity and voice quality. Our simulation experiments demonstrate stability of the algorithm and confirm its self-organizing capability. They also indicate a significant decrease of call blocking and dropping and other quality-of-service improvements.

The shape of illusory figures

We have been developing a stochastic model for figure-ground separation. The model selects/constructs the foreground with preference for figures with “more convex” shapes. When these models are applied to illusory figures they yield perceptually accurate selection of figure and background. The approach is based on an “entropy” measure of a region diffusion Markov model from a set of local figure/ground hypothesis. The contour boundaries are implicitly represented, via the thresholding of the diffusion result. What optimal properties do the illusory contours satisfies? We show that the entropy criteria selects contours such as to minimize a Taylor series of the even derivatives with respect to the length of the contour. The coefficients are positive and they get exponentially smaller as the derivatives increase. The zeroth order term suggest that small length contours are preferred, the second order terms suggests that curvature-like term is minimized (with less strength compared to the zero order one), and higher order derivatives give additional contour smoothness constraints.

Massively parallel simulations of ATM systems

We simulate models of ATM communication systems on a massively parallel SIMD computer. Fast simulations of ATM models are needed because the regimes of interest usually involve high volumes of traffic and low failure rates. Unexpected practical and theoretical difficulties, partly due to the massive parallelism and SIMD aspects, were encountered and we show how to cope with them. In a replica-parallel simulation of an ATM system, large variations in computed statistics are caused by small differences in the distribution of employed random number generators. A comparison of these distributions using a secondary statistical measure served to disambiguate the results. It was also found that time-parallel simulations of ATM systems with Markov sources can be efficiently performed using parallel prefix methods only when the sources have a small number of states, while more complex sources require end-state matching for efficient simulation. We discovered that, with the proper choice of initial state distributions and partial regeneration points, the time and memory requirements can be much improved. Our simulations were carried out on the MasPar MP-1216 system with 16,384 processors, which was compared against an SGI workstation. We achieved about 60%-70% efficiency (speed-up of approx 35 compared to the ideal of approx 51).