Shallow neural networks trained to detect collisions recover features of visual loom-selective neurons

Animals have evolved sophisticated visual circuits to solve a vital inference problem: detecting whether or not a visual signal corresponds to an object on a collision course. Such events are detected by specific circuits sensitive to visual looming, or objects increasing in size. Various computational models have been developed for these circuits, but how the collision-detection inference problem itself shapes the computational structures of these circuits remains unknown. Here, inspired by the distinctive structures of LPLC2 neurons in the visual system of Drosophila, we build anatomically-constrained shallow neural network models and train them to identify visual signals that correspond to impending collisions. Surprisingly, the optimization arrives at two distinct, opposing solutions, only one of which matches the actual dendritic weighting of LPLC2 neurons. Both solutions can solve the inference problem with high accuracy when the population size is large enough. The LPLC2-like solutions reproduces experimentally observed LPLC2 neuron responses for many stimuli, and reproduces canonical tuning of loom sensitive neurons, even though the models are never trained on neural data. Thus, LPLC2 neuron properties and tuning are predicted by optimizing an anatomically-constrained neural network to detect impending collisions. More generally, these results illustrate how optimizing inference tasks that are important for an animal's perceptual goals can reveal and explain computational properties of specific sensory neurons.

Shallow neural networks trained to detect collisions recover features of visual loom-selective neurons

Animals have evolved sophisticated visual circuits to solve a vital inference problem: detecting whether or not a visual signal corresponds to an object on a collision course. Such events are detected by specific circuits sensitive to visual looming, or objects increasing in size. Various computational models have been developed for these circuits, but how the collision-detection inference problem itself shapes the computational structures of these circuits remains unknown. Here, inspired by the distinctive structures of LPLC2 neurons in the visual system of Drosophila, we build an anatomically-constrained shallow neural network model and train it to identify visual signals that correspond to impending collisions. Surprisingly, the optimization arrives at two distinct, opposing solutions, only one of which matches the actual dendritic weighting of LPLC2 neurons. The LPLC2-like solutions are favored when a population of units is trained on the task, but not when units are trained in isolation. The trained model reproduces experimentally observed LPLC2 neuron responses for many stimuli, and reproduces canonical tuning of loom sensitive neurons, even though the model are never trained on neural data. These results show that LPLC2 neuron properties and tuning are predicted by optimizing an anatomically-constrained neural network to detect impending collisions.

Graph Properties of the Adult Drosophila Central Brain

The recent Drosophila central brain connectome offers the possibility of analyzing the graph properties of the fly brain. Crucially, this connectome is dense, meaning all nodes and links are represented, within the limits of experimental error. We consider the connectome as a directed graph with weighted edges. This enables us to look at a number of graph properties, compare them to human designed logic systems, and speculate on how this may affect function. We look at input and output distributions, randomness of wiring, differences between compartments, path lengths, proximity of strong connections, known computational structures, electrical response as a function of compartment structure, and evidence for efficient packing.

Analysis of HMAX Algorithm on Black Bar Image Dataset

An accurate detection and classification of scenes and objects is essential for interacting with the world, both for living beings and for artificial systems. To reproduce this ability, which is so effective in the animal world, numerous computational models have been proposed, frequently based on bioinspired, computational structures. Among these, Hierarchical Max-pooling (HMAX) is probably one of the most important models. HMAX is a recognition model, mimicking the structures and functions of the primate visual cortex. HMAX has already proven its effectiveness and versatility. Nevertheless, its computational structure presents some criticalities, whose impact on the results has never been systematically assessed. Traditional assessments based on photographs force to choose a specific context; the complexity of images makes it difficult to analyze the computational structure. Here we present a new, general and unspecific assessment of HMAX, introducing the Black Bar Image Dataset, a customizable set of images created to be a universal and flexible model of any ‘real’ image. Results: surprisingly, HMAX demonstrates a notable sensitivity also with a low contrast of luminance. Images containing a wider information pattern enhance the performances. The presence of textures improves performance, but only if the parameterization of the Gabor filter allows its correct encoding. In addition, in complex conditions, HMAX demonstrates good effectiveness in classification. Moreover, the present assessment demonstrates the benefits offered by the Black Bar Image Dataset, its modularity and scalability, for the functional investigations of any computational models.

Abstraction and Representational Capacity in Computational Structures

Michael Weisberg’s book Simulation and Similarity argued that although mathematical models are sometimes described in narrative form, they are best understood as interpreted mathematical structures. But how can a mathematical structure be causal, as many models described in narrative seem to be? This chapter argues that models with apparently narrative form are actually computational structures. It explores this suggestion in detail, examining what computational structure consists of, the resources it offers modelers, and why attempting to re-describe computational models as imaginary concrete systems fails even more dramatically than it does for mathematical models.

Gaming Systems

The ubiquitous adoption of mobile computing devices has implicated all of us in a techno-social system of interaction dominated by the codified and computational logic of the game. This paper will examines the modes by which these computational structures, in the guise of games, have come to dominate our understanding of, and interaction with, the non-game world. It will then identify how the application of this logic creates cognitive and phenomenological ruptures, which can be leveraged by creative individuals to reveal logical fallacies within the applied structures. Throughout, it will identify and analyze creative practices that exemplify responses to these logical fallacies in order to identify ways in which a new class of creative individuals is emerging to tackle the dangerous slippage between gamespace (the space of play, games) and gameic (gamic) space (ordinary/real life to which ludic properties have been applied).

A Comparative Study of Two Families of Higher-Order Accurate Time Integration Algorithms

Two families of higher-order accurate time integration algorithms are numerically tested by using various nonlinear problems of structural dynamics, and the numerical results obtained from them are compared. To be specific, the higher-order algorithms of Kim and Reddy and the higher-order algorithms of Fung are used for this study. In linear analyses, these two different families of higher-order algorithms do not present noticeable differences. However, performances of these algorithms are quite different when they are applied to various nonlinear dynamic problems. For the numerical tests, well-known nonlinear problems are selected from the past studies. For the completeness, the two families of algorithms are briefly reviewed, and their advantageous computational structures are also explained.

Computational Structures and SAPT Interaction Energies of HXeSH···H2Y (Y=O or S) Complexes

Ab initio calculations of the structures, vibrational spectra and supermolecular and symmetry-adapted perturbation theory (SAPT) interaction energies of the HXeOH and HXeSH complexes with H2O and H2S molecules are presented. Two minima already reported in the literature were reproduced and ten new ones were found together with some transition states. All complexes show blue shift in Xe–H stretching mode upon complexation. The computed spectra suggest that it should be possible to detect and distinguish the complexes experimentally. The structures where H2O or H2S is the proton-donor were found to be the most stable for all complex compositions. The SAPT analysis shows significant differences between the complexes with H2O and H2S indicating much larger dispersion and exchange contributions in the complexes with H2S.