Observe and research 5083 Alumium spinking

Deep neural networks are proven to be highly effective tools in various domains, yet their computational and memory costs restrict them from being widely deployed on portable devices. The recent rapid increase of edge computing devices has led to an active search for techniques to address the above-mentioned limitations of machine learning frameworks. The quantization of artificial neural networks (ANNs), which converts the full-precision synaptic weights into low-bit versions, emerged as one of the solutions. At the same time, spiking neural networks (SNNs) have become an attractive alternative to conventional ANNs due to their temporal information processing capability, energy efficiency, and high biological plausibility. Despite being driven by the same motivation, the simultaneous utilization of both concepts has not been fully studied. Therefore, this thesis work aims to bridge the gap between recent progress in quantized neural networks and SNNs. It presents an extensive study on the performance of the quantization function, represented as a linear combination of sigmoid functions, exploited in low-bit weight quantization in SNNs. The given quantization function demonstrates the state-of-the-art performance on four popular benchmarks, CIFAR10-DVS, DVS128 Gesture, N-Caltech101, and N-MNIST, for binary networks (64.05%, 95.45%, 68.71%, and 99.365 respectively) with small accuracy drops (8.03%, 1.18%, 3.47%, and 0.17% respectively) and up to 32x memory savings, which outperforms the existing methods.

Yes / Spiking Neural (SN) P systems is a particular class of P systems that abstracts and applies ideas from neurobiology. Various aspects, representations and features have been studied extensively, but the tool support for modelling and analysing such systems is relatively limited. In this paper, we present a methodology that maps some classes of SN P systems to the equivalent kernel P system representations, which allows analysing SN P system dynamics using the kPWORKBENCH tool. We illustrate the applicability of our approach in some case studies, including an example system from synthetic biology.

Spiking neural networks (SNNs) are a more biologically realistic model of the brain than traditional analog neural networks and therefore should be better for modelling certain functions of the human brain. This thesis uses the concept of deriving an SNN from an accepted non-spiking neural network via analysis and modifications of the transmission function. We investigate this process to determine if and how the modifications can be made to minimize loss of information during the transition from non-spiking to spiking while retaining positive features and functionality of the non-spiking network. By comparing combinations of spiking neuron models and networks against each other, we determined that replacing the transmission function with a neural model that is similar to it allows for the easiest method to create a spiking neural network that works comparatively well. This similarity between transmission function and neuron model allows for easier parameter selection which is a key component in getting a functioning SNN. The parameters all play different roles, but for the most part, parameters that speed up spiking, such as large resistance values or small rheobases generally help the accuracy of the network. But the network is still incomplete for a spiking neural network since this conversion is often only performed after learning has been completed in analog form. The neuron model and subsequent network developed here are the initial steps in creating a bidirectional SNN that handles hetero-associative and auto-associative recall and can be switched easily between spiking and non-spiking with minimal to no loss of data. By tying everything to the transmission function, the non-spiking learning rule, which in our case uses the transmission function, and the neural model of the SNN, we are able to create a functioning SNN. Without this similarity, we find that creating SNN are much more complicated and require much more work in parameter optimization to achieve a functioning SNN.

Artificial intelligence is a swiftly growing field, and many are researching whether AI can serve as a diagnostic aid in the medical domain. However, the primary weakness of traditional machine learning for many applications is energy efficiency, and this may hamper its ability to be effectively utilized in medicine for portable or edge systems. In order to be more effective, new energy-efficient machine learning paradigms must be investigated for medical applications. In addition, smaller models with fewer parameters would be better suited to medical edge systems. By processing data as a series of "spikes" instead of continuous values, spiking neural networks (SNN) may be the right model architecture to address these concerns. This work investigates the proposed advantages of SNNs compared to more traditional architectures when tested on various medical datasets. We compare the energy efficiency of SNN and recurrent neural network (RNN) solutions by finding sizes of each architecture that achieve similar accuracy. The energy consumption of each comparable network is assessed using standard tools for such evaluation. On the SEED human emotion dataset, SNN architectures achieved up to 20x lower energy per inference than an RNN while maintaining similar classification accuracy. SNNs also achieved 30x lower energy consumption on the PTB-XL ECG dataset with similar classification accuracy. These results show that spiking neural networks are more energy efficient than traditional machine learning models at inference time while maintaining a similar level of accuracy for various medical classification tasks. With this superior energy efficiency, this makes it possible for medical SNNs to operate on edge and portable systems. / Master of Science / As artificial intelligence grows in popularity, especially with the rise of new large language models like Chat-GPT, a weakness in traditional architectures becomes more pronounced. These AI models require ever-increasing amounts of energy to operate. Thus, there is a need for more energy-efficient AI models, such as the spiking neural network (SNN). In SNNs, information is processed in a series of spiking signals, like the biological brain. This allows the resulting architecture to be highly energy efficient and adapted to processing time-series data. A domain that often encounters time-series data and would benefit from greater energy efficiency is medicine. This work seeks to investigate the proposed advantages of spiking neural networks when applied to the various classification tasks in the medical domain. Specifically, both an SNN and a traditional recurrent neural network (RNN) were trained on medical datasets for brain signal and heart signal classification. Sizes of each architecture were found that achieved similar classification accuracy and the energy consumption of each comparable network was assessed. For the SEED brain signal dataset, the SNN achieved similar classification accuracy to the RNN while consuming as little as 5% of the energy per inference. Similarly, the SNN consumed 30x less energy than the RNN while classifying the PTB-XL ECG dataset. These results show that the SNN architecture is a more energy efficient model than traditional RNNs for various medical tasks at inference time and may serve as the solution to the energy consumption problem of medical AI.

Previous studies have shown that it is possible for juvenile songbirds to learn songs through listening to prerecorded songs played back to them. What is not known however, is how this will differ from normal learning, both on neural level as well as on the bird as whole. In this project we have taken data from playback experiments and attempted to measure the differences in neuron spiking activity across two different contexts. Firstly, when the bird is only listening to playback recordings and secondly when the bird is listening to playback recordings on the same day as listening to a live tutor. We analysed the spiking activity with several different methods in order to establish a distinction between these contexts that could be seen across birds and across trials. The methods include analysing joint spiking events as binary spike trains, the Fano factor across trials as well as the variability of the spike rate. Our hypothesis was that the birds would learn more effectively on days when exposed to a live tutor, therefore, the playback days would show much higher and much more varied spiking data. From our results we found many cases when this hypothesis is true. However, it does not hold true for each of the birds, as some are offered similar results in either context. Therefore we believe that further study would be required to get conclusive results. Although, our results tend to favour the tutoring days it is only suggestive that this shows evidence of better learning. / Tidigare studier har visat att det är möjligt för unga sångfåglar att lära sig sånger genom att lyssna på förinspelade läten som spelas upp för dem. Vad som dock inte är känt är hur detta sätt kommer att skilja sig, jämfört med normalt lärande, både på neural nivå och på fågeln som helhet. I detta projekt har vi tagit data från ett uppspelningsexperiment och försökt mäta skillnaderna neuronspikande aktiviteter i två fall. I det första fallet lyssnar fågeln bara på inspelad fågelsång och i det andra fallet lyssnar fågeln på inspelad fågelsång samma dag som den lyssnar på en vuxen fågel som mentor. Vi analyserade spikningsaktiviteten med flera olika metoder för att finna en skillnad mellan dessa fall, som kan ses både mellan fåglar och mellan ollika försök. Metoderna inkluderar analys av gemensamma spikinghändelser som binära spiktåg, Fanofaktorn över försöken samt variationen i spikhastigheten. Vår hypotes var att fåglarna skulle lära sig mer effektivt på dagar när de utsattes för en vuxen fågel som mentor, därför skulle uppspelningsdagarna visa mycket högre och mycket mer varierad spikdata. Från våra resultat fann vi att i många fall där hypotesen är sann. Men den stämmer inte för alla fåglarna eftersom några fåglar hade liknande resultat för båda fallen. Därför tror vi att ytterligare studier krävs för att få tydliga resultat. Dock så tenderar våra resultat att gynna mentordagarna, även om det bara antyder att detta visar på bättre lärande.

The hippocampus, in humans and rats, plays crucial roles in spatial tasks and nonspatial tasks involving episodic-type memory. This thesis presents a novel computational model of the hippocampus (CA1, CA3 and dentate gyrus) which creates a framework where spatial memory and episodic memory are explained together. This general model follows the approach where the memory function of the rodent hippocampus is seen as a “memory space” instead of a “spatial memory”. The innovations of this novel model are centred around the fact that it follows detailed hippocampal architecture constraints and uses spiking networks to represent all hippocampal subfields. This hippocampal model does not require stable attractor states to produce a robust memory system capable of pattern separation and pattern completion. In this hippocampal theory, information is represented and processed in the form of activity patterns. That is, instead of assuming firing-rate coding, this model assumes that information is coded in the activation of specific constellations of neurons. This coding mechanism, associated with the use of spiking neurons, raises many problems on how information is transferred, processed and stored in the different hippocampal subfields. This thesis explores which mechanisms are available in the hippocampus to achieve such control, and produces a detailed model which is biologically realistic and capable of explaining how several computational components can work together to produce the emergent functional properties of the hippocampus. In this hippocampal theory, precise explanations are given to why mossy fibres are important for storage but not recall, what is the functional role of the mossy cells (excitatory interneurons) in the dentate gyrus, why firing fields can be asymmetric with the firing peak closer to the end of the field, which features are used to produce “place fields”, among others. An important property of this hippocampal model is that the memory system provided by the CA3 is a palimpsest memory: after saturation, the number of patterns that can be recalled is independent of the number of patterns engraved in the recurrent network. In parallel with the development of the hippocampal computational model, a simulation environment was created. This simulation environment was tailored for the needs and assumptions of the hippocampal model and represents an important component of this thesis.

This project applies principles from the field of Computational Neuroscience to Robotics research, in particular to develop systems inspired by how nature manages to solve sensorimotor coordination tasks. The overall aim has been to build a self-organising sensorimotor system using biologically inspired techniques based upon human cortical development which can in the future be implemented in neuromorphic hardware. This can then deliver the benefits of low power consumption and real time operation but with flexible learning onboard autonomous robots. A core principle is the Self-Organising Feature Map which is based upon the theory of how 2D maps develop in real cortex to represent complex information from the environment. A framework for developing feature maps for both motor and visual directional selectivity representing eight different directions of motion is described as well as how they can be coupled together to make a basic visuomotor system. In contrast to many previous works which use artificially generated visual inputs (for example, image sequences of oriented moving bars or mathematically generated Gaussian bars) a novel feature of the current work is that the visual input is generated by a DVS 128 silicon retina camera which is a neuromorphic device and produces spike events in a frame-free way. One of the main contributions of this work has been to develop a method of autonomous regulation of the map development process which adapts the learning dependent upon input activity. The main results show that distinct directionally selective maps for both the motor and visual modalities are produced under a range of experimental scenarios. The adaptive learning process successfully controls the rate of learning in both motor and visual map development and is used to indicate when sufficient patterns have been presented, thus avoiding the need to define in advance the quantity and range of training data. The coupling training experiments show that the visual input learns to modulate the original motor map response, creating a new visual-motor topological map.

Autor Markéta Popelová Název práce Software tool for modelling coding and processing of information in auditory cortex of mice Abstrakt Porozumění zpracovávání a kódování informací ve sluchové k·ře (AC) je stále ne- dostatečné. Z několika r·zných d·vod· by bylo užitečné mít výpočetní model AC, například z d·vodu vysvětlení, či ujasnění procesu kódování informací v AC. Prv- ním cílem této práce bylo vytvořit softwarový nástroj (simulátor SUSNOMAC), zaměřený na modelování AC. Druhým cílem bylo navrhnout výpočetní model AC s následujícími vlastnostmi: Izhikevich·v model neuronu, dlouhodobá plasticita ve formě Spike-timing-dependent plasticity (STDP), šestivrstvá architektura, pa- rametrizované typy neuron·, hustota neuron· a pravděpodobnost vzniku synapsí. Navržený model byl testován v desítkách experiment·, s r·znými sadami para- metr· a v r·zných velikostech (až 100 000 neuron· s takřka 21 milióny synapsí). Experimenty byly analyzovány a jejich výsledky srovnány s pozorováním skutečné AC. V práci popisujeme a analyzujeme několik zajímavých pozorování o aktivitě modelované sítě a vzniku tonotopického uspořádání AC. 1

The current study assessed features associated with drink spiking, or the adding of a substance to another person's drink without the consumer's knowledge or consent. A sample of 805 Australians, aged 18-35 years, completed a survey designed to measure the occurrence and predictors of the perpetration and victimisation of drink spiking. Almost half of the sample reported at least one experience of purchasing or mixing cocktails for others (49% and 45%, respectively), while smaller proportions reported adding alcohol to punch (26%) and adding alcoholic shots to alcoholic beverages belonging to other people (16%). A number of participants also reported previous experience of adding alcoholic shots to non-alcoholic beverages (6%), adding prescription or illicit substances to alcoholic beverages (1%), adding substances to non-alcoholic beverages (1%), and adding substances to punch (1%). Purchasing or mixing cocktails for others, adding alcohol to punch, or adding alcoholic shots to alcoholic beverages were predicted by beliefs that deliberately causing intoxication in others is acceptable and that alcohol consumption by others is indicative of their sexual attraction to participants. Engagement in these behaviours was also predicted by participants' illicit substance use and participation in casual sexual activity. Adding prescription or illicit substances to other people's beverages, or adding alcoholic shots to non-alcoholic beverages, were predicted by the belief that alcohol consumption increases one's confidence and sexual responsiveness, and by participants' use of narcotics and sedatives. Perpetrators were predominantly motivated by a wish to have fun or to increase the likelihood of engaging in consensual sexual activity. With regard to victimisation of drink spiking, 26% of the sample reported at least one victimisation. The majority of incidents occurred in licensed venues, after the participant had engaged in such low supervisory behaviours as leaving their drink unattended or accepting a drink without observing its preparation. Most participants established a belief that they had been spiked after experiencing a degree of intoxication that was beyond their expected level (based on the amount of alcohol consumption), or after experiencing such physiological symptoms as vomiting, hallucinations, lack of coordination, or unconsciousness. Despite such experiences, 85% of victims did not report the incident to authorities. Victimisation in general was predicted by participants' use of stimulant and hallucinogenic substances. Female victimisation was predicted by previous episodes of victimisation of oral sexual assault. Victimisation was not affected by participants' degree of supervision of their drinks. These findings provided empirical evidence that drink spiking is committed primarily for the purposes of creating a fun, entertaining situation. However, it was also apparent that drink spiking is perpetrated in an attempt to encourage participation in consensual sexual activity; this was particularly the case in incidents involving the addition of substances, as opposed to alcohol, to beverages belonging to others. Conclusions regarding the motivations held by perpetrators of drink spiking and the post-spiking experiences of victims informed the provision of recommendations for intervention for victims and prevention programs aimed at reducing the incidence of victimisation in the future.