Photonics-based neural networks promise to outperform electronic counterparts, accelerating neural network computations while reducing power consumption and footprint. However, these solutions suffer from physical layer constraints arising from the underlying analog photonic hardware, impacting the resolution of computations (in terms of effective number of bits), requiring the use of positive-valued inputs, and imposing limitations in the fan-in and in the size of convolutional kernels. To abstract these constraints, in this paper we introduce the concept of Photonic-Aware Neural Network (PANN) architectures, i.e., deep neural network models aware of the photonic hardware constraints. Then, we devise PANN training schemes resorting to quantization strategies aimed to obtain the required neural network parameters in the fixed-point domain, compliant with the limited resolution of the underlying hardware. We finally carry out extensive simulations exploiting PANNs in image classification tasks on well-known datasets (MNIST, Fashion-MNIST, and Cifar-10) with varying bitwidths (i.e., 2, 4, and 6 bits). We consider two kernel sizes and two pooling schemes for each PANN model, exploiting $$2\times 2$$2 × 2 and $$3\times 3$$3 × 3 convolutional kernels, and max and average pooling, the latter more amenable to an optical implementation. $$3\times 3$$3 × 3 kernels perform better than $$2\times 2$$2 × 2 counterparts, while max and average pooling provide comparable results, with the latter performing better on MNIST and Cifar-10. The accuracy degradation due to the photonic hardware constraints is quite limited, especially on MNIST and Fashion-MNIST, demonstrating the feasibility of PANN approaches on computer vision tasks.

Reconfigurable linear optical processors can be used to perform linear transformations and are instrumental in effectively computing matrix–vector multiplications required in each neural network layer. In this paper, we characterize and compare two thermally tuned photonic integrated processors realized in silicon-on-insulator and silicon nitride platforms suited for extracting feature maps in convolutional neural networks. The reduction in bit resolution when crossing the processor is mainly due to optical losses, in the range 2.3–3.3 for the silicon-on-insulator chip and in the range 1.3–2.4 for the silicon nitride chip. However, the lower extinction ratio of Mach–Zehnder elements in the latter platform limits their expressivity (i.e., the capacity to implement any transformation) to 75%, compared to 97% of the former. Finally, the silicon-on-insulator processor outperforms the silicon nitride one in terms of footprint and energy efficiency.

Despite the increased exploration of machine learning (ML) techniques for the realization of autonomous optical networks, less attention has been paid to data quality, which is critical for ML performance. Failure management in optical networks using ML is constrained by the fact that some failures may occur more frequently than others, resulting in highly imbalanced datasets for the training of ML models. To address this limitation, a variational-autoencoder-based data augmentation technique is investigated in this paper, which can be used during data preprocessing to improve data quality. The synthetic data generated by the variational autoencoder are utilized to reduce imbalance in an experimental dataset used for training of neural networks (NNs) for failure management in optical networks. First, it is shown that, with a modified training dataset, the training time of NNs can be reduced. Reductions of up to 37.1% and 60.6% are achieved for failure detection and cause identification, respectively. Second, it is shown that improvement in the quality of the training dataset can reduce the computational complexity of NNs during the inference phase. As determined analytically, almost 68% reduction in computational complexity is achieved for the NN used for failure cause identification. Finally, data augmentation is shown to achieve improvement in classification accuracy. This work demonstrates improvement of up to 7.32%.

In this paper, we propose a methodology to maximize the benefits of interdisciplinary cooperation in AI research groups. Firstly, we build the case for the importance of interdisciplinarity in research groups as the best means to tackle the social implications brought about by AI systems, against the backdrop of the EU Commission proposal for an Artificial Intelligence Act. As we are an interdisciplinary group, we address the multi-faceted implications of the mass-scale diffusion of AI-driven technologies. The result of our exercise lead us to postulate the necessity of a behavioural theory that standardizes the interaction process of interdisciplinary groups. In light of this, we conduct a review of the existing approaches to interdisciplinary research on AI appliances, leading to the development of methodologies like ethics-by-design and value-sensitive design, evaluating their strengths and weaknesses. We then put forth an iterative process theory hinging on a narrative approach consisting of four phases: (i) definition of the hypothesis space, (ii) building-up of a common lexicon, (iii) scenario-building, (iv) interdisciplinary self-assessment. Finally, we identify the most relevant fields of application for such a methodology and discuss possible case studies.