LAF-Net: Locally Adaptive Fusion Networks for Stereo Confidence Estimation

We present a novel method that estimates confidence map of an initial disparity by making full use of tri-modal input, including matching cost, disparity, and color image through deep networks. The proposed network, termed as Locally Adaptive Fusion Networks (LAF-Net), learns locally-varying attention and scale maps to fuse the trimodal confidence features. The attention inference networks encode the importance of tri-modal confidence features and then concatenate them using the attention maps in an adaptive and dynamic fashion. This enables us to make an optimal fusion of the heterogeneous features, compared to a simple concatenation technique that is commonly used in conventional approaches. In addition, to encode the confidence features with locally-varying receptive fields, the scale inference networks learn the scale map and warp the fused confidence features through convolutional spatial transformer networks. Finally, the confidence map is progressively estimated in the recursive refinement networks to enforce a spatial context and local consistency. Experimental results show that this model outperforms the state-ofthe-art methods on various benchmarks.

Geometry-Aware Deep Recurrent Neural Networks for Hyperspectral Image Classification

Siyuan Hao, Mathieu Salzmann, Wei Wang

Variants of deep networks have been widely used for hyperspectral image (HSI)-classification tasks. Among them, in recent years, recurrent neural networks (RNNs) have attracted considerable attention in the remote sensing community. However, complex geometries cannot be learned easily by the traditional recurrent units [e.g., long short-term memory (LSTM) and gated recurrent unit (GRU)]. In this article, we propose a geometry-aware deep recurrent neural network (Geo-DRNN) for HSI classification. We build this network upon two modules: a U-shaped network (U-Net) and RNNs. We first input the original HSI patches to the U-Net, which can be trained with very few images and obtain a preliminary classification result. We then add RNNs on the top of the U-Net so as to mimic the human brain to refine continuously the output-classification map. However, instead of using the traditional dot product in each gate of the RNNs, we introduce a Net-Gated GRU that increases the nonlinear representation power. Finally, we use a pretrained ResNet as a regularizer to improve further the ability of the proposed network to describe complex geometries. To this end, we construct a geometry-aware ResNet loss, which leverages the pretrained ResNet's knowledge about the different structures in the real world. Our experimental results on real HSIs and road topology images demonstrate that our approach outperforms the state-of-the-art classification methods and can learn complex geometries.

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC2021

Evolutionary and Computational Advantages of Neuromodulated Plasticity

Andrea Soltoggio

The integration of modulatory neurons into evolutionary artificial neural networks is proposed here. A model of modulatory neurons was devised to describe a plasticity mechanism at the low level of synapses and neurons. No initial assumptions were made on the network structures or on the system level dynamics. The work of this thesis studied the outset of high level system dynamics that emerged employing the low level mechanism of neuromodulated plasticity. Fully-fledged control networks were designed by simulated evolution: an evolutionary algorithm could evolve networks with arbitrary size and topology using standard and modulatory neurons as building blocks. A set of dynamic, reward-based environments was implemented with the purpose of eliciting the outset of learning and memory in networks. The evolutionary time and the performance of solutions were compared for networks that could or could not use modulatory neurons. The experimental results demonstrated that modulatory neurons provide an evolutionary advantage that increases with the complexity of the control problem. Networks with modulatory neurons were also observed to evolve alternative neural control structures with respect to networks without neuromodulation. Different network topologies were observed to lead to a computational advantage such as faster input-output signal processing. The evolutionary and computational advantages induced by modulatory neurons strongly suggest the important role of neuromodulated plasticity for the evolution of networks that require temporal neural dynamics, adaptivity and memory functions.

University of Birmingham, UK2008

Cortical Dynamics in Presence of Assemblies of Densely Connected Weight-Hub Neurons

Moritz Deger, Wulfram Gerstner, Hesam Setareh

Experimental measurements of pairwise connection probability of pyramidal neurons together with the distribution of synaptic weights have been used to construct randomly connected model networks. However, several experimental studies suggest that both wiring and synaptic weight structure between neurons show statistics that differ from random networks. Here we study a network containing a subset of neurons which we call weight-hub neurons, that are characterized by strong inward synapses. We propose a connectivity structure for excitatory neurons that contain assemblies of densely connected weight-hub neurons, while the pairwise connection probability and synaptic weight distribution remain consistent with experimental data. Simulations of such a network with generalized integrate-and-fire neurons display regular and irregular slow oscillations akin to experimentally observed up/down state transitions in the activity of cortical neurons with a broad distribution of pairwise spike correlations. Moreover, stimulation of a model network in the presence or absence of assembly structure exhibits responses similar to light-evoked responses of cortical layers in optogenetically modified animals. We conclude that a high connection probability into and within assemblies of excitatory weight-hub neurons, as it likely is present in some but not all cortical layers, changes the dynamics of a layer of cortical microcircuitry significantly.

Frontiers Research Foundation2017