Temporal Coding in Spiking Neural Networks With Alpha Synaptic Function: Learning With Backpropagation.

The timing of individual neuronal spikes is essential for biological brains to make fast responses to sensory stimuli. However, conventional artificial neural networks lack the intrinsic temporal coding ability present in biological networks. We propose a spiking neural network model that encodes information in the relative timing of individual spikes. In classification tasks, the output of the network is indicated by the first neuron to spike in the output layer. This temporal coding scheme allows the supervised training of the network with backpropagation, using locally exact derivatives of the postsynaptic spike times with respect to presynaptic spike times. The network operates using a biologically plausible synaptic transfer function. In addition, we use trainable pulses that provide bias, add flexibility during training, and exploit the decayed part of the synaptic function. We show that such networks can be successfully trained on multiple data sets encoded in time, including MNIST. Our model outperforms comparable spiking models on MNIST and achieves similar quality to fully connected conventional networks with the same architecture. The spiking network spontaneously discovers two operating modes, mirroring the accuracy-speed tradeoff observed in human decision-making: a highly accurate but slow regime, and a fast but slightly lower accuracy regime. These results demonstrate the computational power of spiking networks with biological characteristics that encode information in the timing of individual neurons. By studying temporal coding in spiking networks, we aim to create building blocks toward energy-efficient, state-based biologically inspired neural architectures. We provide open-source code for the model.

Spiking Autoencoders With Temporal Coding.

Spiking neural networks with temporal coding schemes process information based on the relative timing of neuronal spikes. In supervised learning tasks, temporal coding allows learning through backpropagation with exact derivatives, and achieves accuracies on par with conventional artificial neural networks. Here we introduce spiking autoencoders with temporal coding and pulses, trained using backpropagation to store and reconstruct images with high fidelity from compact representations. We show that spiking autoencoders with a single layer are able to effectively represent and reconstruct images from the neuromorphically-encoded MNIST and FMNIST datasets. We explore the effect of different spike time target latencies, data noise levels and embedding sizes, as well as the classification performance from the embeddings. The spiking autoencoders achieve results similar to or better than conventional non-spiking autoencoders. We find that inhibition is essential in the functioning of the spiking autoencoders, particularly when the input needs to be memorised for a longer time before the expected output spike times. To reconstruct images with a high target latency, the network learns to accumulate negative evidence and to use the pulses as excitatory triggers for producing the output spikes at the required times. Our results highlight the potential of spiking autoencoders as building blocks for more complex biologically-inspired architectures. We also provide open-source code for the model.

Testing of the analytical anisotropic algorithm for photon dose calculation.

The analytical anisotropic algorithm (AAA) was implemented in the Eclipse (Varian Medical Systems) treatment planning system to replace the single pencil beam (SPB) algorithm for the calculation of dose distributions for photon beams. AAA was developed to improve the dose calculation accuracy, especially in heterogeneous media. The total dose deposition is calculated as the superposition of the dose deposited by two photon sources (primary and secondary) and by an electron contamination source. The photon dose is calculated as a three-dimensional convolution of Monte-Carlo precalculated scatter kernels, scaled according to the electron density matrix. For the configuration of AAA, an optimization algorithm determines the parameters characterizing the multiple source model by optimizing the agreement between the calculated and measured depth dose curves and profiles for the basic beam data. We have combined the acceptance tests obtained in three different departments for 6, 15, and 18 MV photon beams. The accuracy of AAA was tested for different field sizes (symmetric and asymmetric) for open fields, wedged fields, and static and dynamic multileaf collimation fields. Depth dose behavior at different source-to-phantom distances was investigated. Measurements were performed on homogeneous, water equivalent phantoms, on simple phantoms containing cork inhomogeneities, and on the thorax of an anthropomorphic phantom. Comparisons were made among measurements, AAA, and SPB calculations. The optimization procedure for the configuration of the algorithm was successful in reproducing the basic beam data with an overall accuracy of 3%, 1 mm in the build-up region, and 1%, 1 mm elsewhere. Testing of the algorithm in more clinical setups showed comparable results for depth dose curves, profiles, and monitor units of symmetric open and wedged beams below dmax. The electron contamination model was found to be suboptimal to model the dose around dmax, especially for physical wedges at smaller source to phantom distances. For the asymmetric field verification, absolute dose difference of up to 4% were observed for the most extreme asymmetries. Compared to the SPB, the penumbra modeling is considerably improved (1%, 1 mm). At the interface between solid water and cork, profiles show a better agreement with AAA. Depth dose curves in the cork are substantially better with AAA than with SPB. Improvements are more pronounced for 18 MV than for 6 MV. Point dose measurements in the thoracic phantom are mostly within 5%. In general, we can conclude that, compared to SPB, AAA improves the accuracy of dose calculations. Particular progress was made with respect to the penumbra and low dose regions. In heterogeneous materials, improvements are substantial and more pronounced for high (18 MV) than for low (6 MV) energies.

Evidence for a functional role of GABA receptors in the rat mature hippocampus.

Both gamma-aminobutyric acid (GABA)(C) receptor subunit mRNA and protein are expressed in the stratum pyramidale in the CA1 area of the adult rat hippocampus, but so far no conclusive evidence about functional hippocampal GABA(C) receptors has been presented. Here, the contribution of GABA(C) receptors to stimulus-evoked postsynaptic potentials was studied in the hippocampal CA1 area with extracellular and intracellular recordings at the age range of 21-47 postnatal days. Activation of GABA(C) receptors with the specific agonist cis-4-aminocrotonic acid (CACA) suppressed postsynaptic excitability and increased the membrane conductance. The GABA(C) receptor antagonist 1,2,5,6-tetrahydropyridine-4-ylmethylphosphinic acid (TPMPA), but not the GABA(A) receptor antagonist bicuculline, inhibited the effects of CACA. GABA-mediated long-lasting depolarizing responses evoked by high-frequency stimulation of local inhibitory interneurons in the CA1 area in the presence of ionotropic glutamate receptor and GABA(B) receptor blockers were prolonged by TPMPA, indicating that GABA(C) receptors are activated under these conditions. For weaker stimulation, the effect of TPMPA was enhanced after GABA uptake was inhibited. Our data demonstrate that GABA(C) receptors can be activated by endogenous synaptic transmitter release following strong stimulation or under conditions of reduced GABA uptake. The lack of GABA(C) receptor activation by less intensive stimulation under control conditions suggests that these receptors are extrasynaptic and activated via spillover of synaptically released GABA.