This volume comprises the Late-Breaking Abstracts accepted for the Evo* 2023 Conference, hosted in Brno (Czech Republic), from April 12th to 14th. These abstracts were featured in both short talks and the conference's poster session, offering insights into ongoing research and preliminary findings exploring the application of various Evolutionary Computation approaches and other Nature-Inspired techniques to real-world problems. These contributions represent promising developments, highlighting forthcoming advances and applications in the field of nature-inspired methods, particularly Evolutionary Algorithms.

Transformers have revolutionized the machine learning landscape, gradually making their way into everyday tasks and equipping our computers with ``sparks of intelligence''. However, their runtime requirements have prevented them from being broadly deployed on mobile. As personal devices become increasingly powerful and prompt privacy becomes an ever more pressing issue, we explore the current state of mobile execution of Large Language Models (LLMs). To achieve this, we have created our own automation infrastructure, MELT, which supports the headless execution and benchmarking of LLMs on device, supporting different models, devices and frameworks, including Android, iOS and Nvidia Jetson devices. We evaluate popular instruction fine-tuned LLMs and leverage different frameworks to measure their end-to-end and granular performance, tracing their memory and energy requirements along the way. Our analysis is the first systematic study of on-device LLM execution, quantifying performance, energy efficiency and accuracy across various state-of-the-art models and showcases the state of on-device intelligence in the era of hyperscale models. Results highlight the performance heterogeneity across targets and corroborates that LLM inference is largely memory-bound. Quantization drastically reduces memory requirements and renders execution viable, but at a non-negligible accuracy cost. Drawing from its energy footprint and thermal behavior, the continuous execution of LLMs remains elusive, as both factors negatively affect user experience. Last, our experience shows that the ecosystem is still in its infancy, and algorithmic as well as hardware breakthroughs can significantly shift the execution cost. We expect NPU acceleration, and framework-hardware co-design to be the biggest bet towards efficient standalone execution, with the alternative of offloading tailored towards edge deployments.

Neural networks are widely used in all kinds of data processing, especially on seemingly unfeasible tasks such as image [15] and language recognition [10], as well as applications in medicine [16], and prediction of stock markets [6], just to mention a few. Khan et al. [14] provide a survey of such applications, a mathematically oriented textbook concerning structural issues related to Deep Neural Networks is provided by [3]. Neural networks are nowadays also made use of in safety-critical systems like autonomous driving [8] or power grid management. In such a setting, when security issues become important, aspects of certification come into play [7, 11, 17]. If we for example want provable guarantees for certain scenarios to be unreachable, we first need to formulate them as constraints and precisely state for which property of a network we want verification. In the present paper we are interested in studying certain verification problems for NNs in form of particular robustness and minimization problems such as: How will a network react to a small perturbation of the input [9]? And how likely is a network to change the classification of an input that is altered a little? These probabilities are crucial when for example a self-driving car is supposed to recognize a speed limit, and they have already been tackled in practical settings by simulations and heuristic algorithms.

U.S. Ambassador to Japan Rahm Emanuel visited a Fukushima coastal city to support the local fishing industry after China and South Korea raised the alarm over water discharge began from the Fukushima Daiichi nuclear plant. Images taken by miniature drones from deep inside a badly damaged reactor at the Fukushima nuclear plant show displaced control equipment and misshapen materials but leave many questions unanswered, underscoring the daunting task of decommissioning the plant. The 12 photos released by the plant's operator are the first from inside the main structural support called the pedestal in the hardest-hit No. 1 reactor's primary containment vessel, an area directly under the reactor's core. Officials had long hoped to reach the area to examine the core and melted nuclear fuel which dripped there when the plant's cooling systems were damaged by a massive earthquake and tsunami in 2011. Earlier attempts with robots were unable to reach the area.

Eerie new drone footage has for the first time revealed the extent of the damage to the Fukushima nuclear power plant 13 years on from its meltdown. The plant's operators, Tokyo Electric Power Company Holdings, or TEPCO, released 12 photos from inside the site, which are the first ever images from inside the main structural support called the pedestal in the hardest-hit reactor's primary containment vessel, an area directly under the reactor's core. Officials had long hoped to reach the area to examine the core and melted nuclear fuel which dripped there when the plant's cooling systems were damaged by a massive earthquake and tsunami in 2011. The high-definition color images captured by the drones show brown objects with various shapes and sizes dangling from various locations in the pedestal. Parts of the control-rod drive mechanism, which controls the nuclear chain reaction, and other equipment attached to the core were dislodged by the drones. The Fukushima disaster was one of the world's most devastating nuclear mishaps The plant's operators, Tokyo Electric Power Company Holdings, or TEPCO, released 12 photos from inside the site TEPCO officials said they were unable to tell from the images whether the dangling lumps were melted fuel or melted equipment without obtaining other data such as radiation levels.

Nvidia has unveiled a "superchip" for training artificial intelligence models, the most powerful it has ever produced. The US computing firm, which has recently rocketed in value to become the world's third-largest company, has not yet revealed the cost of its new chips, but observers expect a high price tag that will make them accessible to only a few organisations. The chips were announced by Nvidia CEO Jensen Huang at a press conference in San Jose, California on 18 March. He showed off the company's new Blackwell B200 graphics processing units (GPUs), each of which has 208 billion transistors – the tiny switches at the heart of modern computing devices – compared to the 80 billion transistors of Nvidia's current-generation Hopper chips. He also revealed the GB200 Grace Blackwell Superchip, which combines two of the B200 chips.

Faulty wind turbine blades can incur huge costs for the companies that operate them, especially if the defects go unnoticed until it's too late. That's why quality assurance is such a strategic issue for global wind-turbine manufacturers. Today, quality inspections are limited to surface inspection of limited areas as these composite structures roll off the production line. But under a new approach co-created by EPFL and University of Glasgow researchers, inspection engineers can use a new patented radar technology, combined with an AI assistant, to detect possible anomalies beneath the surface. This approach has many advantages: it's non-destructive, non-contact, supports agile and rapid data acquisition and analysis, and requires very little power to operate.

Event cameras, with their high dynamic range and temporal resolution, are ideally suited for object detection, especially under scenarios with motion blur and challenging lighting conditions. However, while most existing approaches prioritize optimizing spatiotemporal representations with advanced detection backbones and early aggregation functions, the crucial issue of adaptive event sampling remains largely unaddressed. Spiking Neural Networks (SNNs), which operate on an event-driven paradigm through sparse spike communication, emerge as a natural fit for addressing this challenge. In this study, we discover that the neural dynamics of spiking neurons align closely with the behavior of an ideal temporal event sampler. Motivated by this insight, we propose a novel adaptive sampling module that leverages recurrent convolutional SNNs enhanced with temporal memory, facilitating a fully end-to-end learnable framework for event-based detection. Additionally, we introduce Residual Potential Dropout (RPD) and Spike-Aware Training (SAT) to regulate potential distribution and address performance degradation encountered in spike-based sampling modules. Through rigorous testing on neuromorphic datasets for event-based detection, our approach demonstrably surpasses existing state-of-the-art spike-based methods, achieving superior performance with significantly fewer parameters and time steps. For instance, our method achieves a 4.4\% mAP improvement on the Gen1 dataset, while requiring 38\% fewer parameters and three time steps. Moreover, the applicability and effectiveness of our adaptive sampling methodology extend beyond SNNs, as demonstrated through further validation on conventional non-spiking detection models.

We study the convergence of gradient flow for the training of deep neural networks. If Residual Neural Networks are a popular example of very deep architectures, their training constitutes a challenging optimization problem due notably to the non-convexity and the non-coercivity of the objective. Yet, in applications, those tasks are successfully solved by simple optimization algorithms such as gradient descent. To better understand this phenomenon, we focus here on a ``mean-field'' model of infinitely deep and arbitrarily wide ResNet, parameterized by probability measures over the product set of layers and parameters and with constant marginal on the set of layers. Indeed, in the case of shallow neural networks, mean field models have proven to benefit from simplified loss-landscapes and good theoretical guarantees when trained with gradient flow for the Wasserstein metric on the set of probability measures. Motivated by this approach, we propose to train our model with gradient flow w.r.t. the conditional Optimal Transport distance: a restriction of the classical Wasserstein distance which enforces our marginal condition. Relying on the theory of gradient flows in metric spaces we first show the well-posedness of the gradient flow equation and its consistency with the training of ResNets at finite width. Performing a local Polyak-\L{}ojasiewicz analysis, we then show convergence of the gradient flow for well-chosen initializations: if the number of features is finite but sufficiently large and the risk is sufficiently small at initialization, the gradient flow converges towards a global minimizer. This is the first result of this type for infinitely deep and arbitrarily wide ResNets.

Photonic integrated circuits play an important role in the field of optical computing, promising faster and more energy-efficient operations compared to their digital counterparts. This advantage stems from the inherent suitability of optical signals to carry out matrix multiplication. However, even deterministic phenomena such as thermal crosstalk make precise programming of photonic chips a challenging task. Here, we train and experimentally evaluate three models incorporating varying degrees of physics intuition to predict the effect of thermal crosstalk in different locations of an integrated programmable photonic mesh. We quantify the effect of thermal crosstalk by the resonance wavelength shift in the power spectrum of a microring resonator implemented in the chip, achieving modelling errors <0.5 pm. We experimentally validate the models through compensation of the crosstalk-induced wavelength shift. Finally, we evaluate the generalization capabilities of one of the models by employing it to predict and compensate for the effect of thermal crosstalk for parts of the chip it was not trained on, revealing root-mean-square-errors of <2.0 pm.