Dr Alex Rast

Implementing in-house AI in the modern business is a classic example of digital transformation, often appearing simple and attractive, particularly given the emergence and availability of powerful, easy-to-use frameworks like TensorFlow or PyTorch. Such AIs are commonly considered for replacing cumbersome manual or physical systems, where neural networks may appear to be almost a panacæa automation solution to solve scalability or diversification concerns. However, such systems have subtle and sometimes very surprising behaviours that require considerable domain expertise, in order to implement a functional system without expending more effort than the system ultimately gains. Fundamentally, they need to be deployed with a clear sense of what the AI system is going to achieve. Careful attention must be paid at the outset to draft a clear and concrete design specification that indicates the intended function, and equally, draws a line under capabilities that are out of scope. Likewise, an effort needs to be made either to identify in-house people with the required skill sets to develop the system, or alternatively to enter into close working partnerships with external providers who can identify the needs and clearly articulate an appropriate solution. Most challenging of all, especially at large scale, is the emerging 'data gap' - the need to have access to or generate enormous volumes of labelled data - which often comes only at costs outside the budget of all but the largest companies. A case study in design collaboration between an emerging company transitioning from a physical to a virtual technology, and a university research group with substantial expertise in AI systems is presented, both as an illustration of the complex design considerations and a model for how to build in-house expertise. The collaboration is ongoing and outcomes are still preliminary, but the company is now starting to gain an appreciation for the complexity of real-world AI deployments and has developed a strategic plan that enables future growth. The emerging overall message is that modern AI is more an exercise in data automation than process automation.

Integrated Circuit (IC) verification, i.e. the process of ensuring that it performs according to the design specifications, is highly resource intensive. This often depends on an IC’s complexity, e.g. the number of gates/transistors which translates into expressions, branches, blocks, etc in its high level descriptions. To reduce overall verification and hence design time, industries resort to “Regression Testing” where a very small test suite, with very high test coverage, is selected to verify any modified design block and its dependencies. One of the key steps in regression test-based verification is distributing the tests to the various interconnected blocks under tests based on their functionality and accessibility, which translates into a block’s “connection strengths” among other parameters. The existing approaches currently define the connection strengths manually by the design experts which often lead to inconsistency in the test results. In this paper, we propose a Graph Neural Network (GNN) based approach to estimate the connection strengths of different interconnected blocks and evaluate its effectiveness with industrial designs in conjunction with a technique called “SMART Regression” compared to random and full regression
testing.

ct from ‘classical’ continuous-valued networks such as multilayer perceptrons (MLPs). With event-driven dynamics and a continuoustime model, in contrast to the discrete-time model of their classical counterparts, they offer interesting advantages in representational capacity and energy consumption. However, developing models of learning for SNNs has historically proven challenging: as continuous-time systems, their dynamics are much more complex and they cannot benefit from the strong theoretical developments in MLPs such as convergence proofs and optimal gradient descent. Nor do they gain automatically from algorithmic improvements that have produced efficient matrix inversion and batch training methods. Research has focussed largely on the most extensively studied learning mechanisms in SNNs: spike-timing-dependent plasticity (STDP). Although there has been progress here, there are also notable pathologies that have often been solved with a variety of ad-hoc techniques. A relatively recent interesting development is attempts to map classical convolutional neural networks to spiking implementations, but these may not leverage all the claimed advantages of spiking. This tutorial overview looks at existing techniques for learning in SNNs and offers some thoughts for future directions.

Biological neurons communicate with each other using two broad categories of spike event coding: rate-based and temporal. Rate-based coding communicates analog information on a continuous scale through the intensity of bursts of spikes while temporal coding relies on the timing of spike events. It has been shown that temporal coding has higher information capacity than rate based coding, but is much more challenging to model due to difficulties estimating spike-time statistics. In this paper we demonstrate how historically dependent NMDA-modulated `resonant' synapses organised in `functional synaptic clusters' provide a robust mechanism for decoding temporally structured spike trains.

In an autonomous driving system, perception - identification of features and objects from the environment
- is crucial. Autonomous racing, in particular, features high speeds and small margins that demand rapid and accurate perception systems. During the race, the weather can change abruptly, causing significant degradation in perception, resulting in ineffective manoeuvres. In order to improve detection in adverse weather, deep-learning-based models typically require extensive datasets captured in such conditions - the
collection of which is a protracted and costly process. However, recent developments in Cycle-GAN architectures allow the synthesis of highly realistic scenes in multiple weather conditions. To this end, we introduce an approach of using synthesised adverse condition datasets in autonomous racing (generated using CycleGAN) to improve the performance of four out of five state-of-the-art detectors by an average of 42.7 and 4.4 mean average precision (mAP) percentage points in the presence of night-time conditions and droplets, respectively. Furthermore, we present a comparative analysis of five object detectors - identifying
the optimal pairing of detector and training data for use during autonomous racing in challenging conditions.

Neuromorphic computing is gaining momentum as an alternative hardware platform for large-scale neural simulation. However, with several major devices and systems available and planned, often with very different characteristics, it is not always clear which platform is suitable for which application. Simulating the platform on conventional computers is typically too slow to be of use, but an alternative approach is to implement an ‘emulation’ of the hardware in FPGAs which can execute at near-hardware speeds but does not commit to a specific hardware architecture. We present an overlay model - a method which superimposes bespoke features on top of a standard template - in both hardware and software to implement neuromorphic architectures using the POETS (Partially Ordered Event Triggered Systems) system. This combination of overlays permits very large-scale simulations to be performed in real time for hardware exploration or application verification, while retaining the flexibility to redefine either the hardware or software layer, if
results indicate potential to improve performance, or significant design problems. Using this system we simulate up to 500,000 neurons on a single-box system, that can be scaled to ∼4,000,000 neurons in an 8-box configuration. Results indicate the crucial constraint for real-time simulation: peak input spike rate per neuron; and help to optimise both hardware and software around neural application requirements. The preliminary architecture demonstrates the feasibility of an overlay model, while indicating directions for future neuromorphic systems. With POETS, we introduce a platform that can help to shape and investigate the neuromorphic architectures of the future.