Deep learning has been successful in various domains including image recognition, speech recognition and natural language processing. However, the research on its application in graph mining is still in an early stage. Here we present Model R, a neural network model created to provide a deep learning approach to the link weight prediction problem. This model uses a node embedding technique that extracts node embeddings (knowledge of nodes) from the known links' weights (relations between nodes) and uses this knowledge to predict the unknown links' weights. We demonstrate the power of Model R through experiments and compare it with the stochastic block model and its derivatives. Model R shows that deep learning can be successfully applied to link weight prediction and it outperforms stochastic block model and its derivatives by up to 73% in terms of prediction accuracy. We analyze the node embeddings to confirm that closeness in embedding space correlates with stronger relationships as measured by the link weight. We anticipate this new approach will provide effective solutions to more graph mining tasks.

Purpose.Retinal implants use electrical stimulation to elicit flashes of light (“phosphenes”). Single-electrode phosphene shape has been shown to vary systematically with stimulus amplitude and frequency as well as the retinal location of the stimulating electrode, due to incidental activation of passing nerve fiber bundles. However, this knowledge has yet to be extended to paired-electrode stimulation.Methods.We retrospectively analyzed 4402 phosphene drawings made by three blind subjects implanted with an Argus II Retinal Prosthesis. Phosphene shape (characterized by area, perimeter, major and minor axis length; normalized per subject) and number of perceived phosphenes were averaged across trials and correlated with the corresponding single-electrode parameters. In addition, the number of phosphenes was correlated with stimulus amplitude and neuroanatomical parameters: electrode-retina (“height”) and electrode-fovea distance (“eccentricity”) as well as the electrode-electrode distance to (“between-axon”) and along axon bundles (“along- axon”). Statistical analyses were conducted using linear regression and partial correlation analysis.Results.Simple regression revealed that each paired-electrode shape descriptor could be predicted by the sum of the two corresponding single- electrode shape descriptors (p < .001). Multiple regression revealed that paired- electrode phosphene shape was primarily predicted by stimulus amplitude, electrode- retina distance, and electrode-fovea distance (p < .05). Interestingly, the number of elicited phosphenes increased with between-axon distance (β= .162, p < .05), but not with along-axon distance (p > .05).Conclusions.The shape of phosphenes elicited by paired-electrode stimulation was well predicted by the shape of their corresponding single-electrode phosphenes, suggesting that two-point perception can be expressed as the linear summation of single-point perception. We also found that the number of perceived phosphenes increased with the between-axon distance of the two electrodes, providing further evidence in support of the axon map model for epiretinal stimulation. These findings contribute to the growing literature on phosphene perception and have important implications for the design of future retinal prostheses.

No abstract

Despite their immense success as a model of macaque visual cortex, deep convolutional neural networks (CNNs) have struggled to predict activity in visual cortex of the mouse, which is thought to be strongly dependent on the animal's behavioral state. Furthermore, most computational models focus on predicting neural responses to static images presented under head fixation, which are dramatically different from the dynamic, continuous visual stimuli that arise during movement in the real world. Consequently, it is still unknown how natural visual input and different behavioral variables may integrate over time to generate responses in primary visual cortex (V1). To address this, we introduce a multimodal recurrent neural network that integrates gaze-contingent visual input with behavioral and temporal dynamics to explain V1 activity in freely moving mice. We show that the model achieves state-of-the-art predictions of V1 activity during free exploration and demonstrate the importance of each component in an extensive ablation study. Analyzing our model using maximally activating stimuli and saliency maps, we reveal new insights into cortical function, including the prevalence of mixed selectivity for behavioral variables in mouse V1. In summary, our model offers a comprehensive deep-learning framework for exploring the computational principles underlying V1 neurons in freely-moving animals engaged in natural behavior.

No abstract

Abstract. For the core of underwater low-light camera, i.e. EMCCD, the whole solution of driving system is presented in this paper. The timing signals which meet the driving requirements of EMCCD are produced from FPGA. The high-speed integrated driver chips convert the timing signals into general power driving signals with the amplitude under 12V; a Class A push-pull amplifier circuit built in discrete components transforms the timing signals into high-voltage multiplying signals. Besides, the impedance matching for the circuit optimizes the driving system. The experimental results indicate that the driving system generates the high-voltage multiplying signals, with their high level adjusted from 30V to 48V and frequency raised up to 10MHz. The whole driving system satisfies the requirements of EMCCD, and has the capability to ensure the normal operation of EMCCD.

No abstract