Abstract
Extreme Learning Machines (ELMs) are a versatile Machine Learning (ML) algorithm that features as the main advantage the possibility of a seamless implementation with physical systems. Yet, despite the success of the physical implementations of ELMs, there is still a lack of fundamental understanding in regard to their optical implementations. In this context, this work makes use of an optical complex media and wavefront shaping techniques to implement a versatile optical ELM playground to gain a deeper insight into these machines. In particular, we present experimental evidences on the correlation between the effective dimensionality of the hidden space and its generalization capability, thus bringing the inner workings of optical ELMs under a new light and opening paths toward future technological implementations of similar principles.

Abstract
Extreme learning machines (ELMs) are a versatile machine learning technique that can be seamlessly implemented with optical systems. In short, they can be described as a network of hidden neurons with random fixed weights and biases, that generate a complex behaviour in response to an input. Yet, despite the success of the physical implementations of ELMs, there is still a lack of fundamental understanding about their optical implementations. This work makes use of an optical complex media to implement an ELM and introduce an ab-initio theoretical framework to support the experimental implementation. We validate the proposed framework, in particular, by exploring the correlation between the rank of the outputs, H, and its generalization capability, thus shedding new light into the inner workings of optical ELMs and opening paths towards future technological implementations of similar principles.

Abstract
Reservoir computing is a versatile approach for implementing physically Recurrent Neural networks which take advantage of a reservoir, consisting of a set of interconnected neurons with temporal dynamics, whose weights and biases are fixed and do not need to be optimized. Instead, the training takes place only at the output layer towards a specific task. One important requirement for these systems to work is nonlinearity, which in optical setups is usually obtained via the saturation of the detection device. In this work, we explore a distinct approach using a photorefractive crystal as the source of the nonlinearity in the reservoir. Furthermore, by leveraging on the time response of the photorefractive media, one can also have the temporal interaction required for such architecture. If we space out in time the propagation of different states, the temporal interaction is lost, and the system can work as an extreme learning machine. This corresponds to a physical implementation of a Feed-Forward Neural Network with a single hidden layer and fixed random weights and biases. Some preliminary results are presented and discussed. © Published under licence by IOP Publishing Ltd.

Abstract
The ability to select, isolate, and manipulate micron-sized particles or small clusters has made optical tweezers one of the emergent tools for modern biotechnology. In conventional setups, the classification of the trapped specimen is usually achieved through the acquired image, the scattered signal, or additional information such as Raman spectroscopy. In this work, we propose a solution that uses the temporal data signal from the scattering process of the trapping laser, acquired with a quadrant photodetector. Our methodology rests on a pre-processing strategy that combines Fourier transform and principal component analysis to reduce the dimension of the data and perform relevant feature extraction. Testing a wide range of standard machine learning algorithms, it is shown that this methodology allows achieving accuracy performances around 90%, validating the concept of using the temporal dynamics of the scattering signal for the classification task. Achieved with 500 millisecond signals and leveraging on methods of low computational footprint, the results presented pave the way for the deployment of alternative and faster classification methodologies in optical trapping technologies.

Abstract
Background: In view of the growing importance of nanotechnologies, the detection/identification of nanoparticles type has been considered of utmost importance. Although the characterization of synthetic/organic nanoparticles is currently considered a priority (eg, drug delivery devices, nanotextiles, theranostic nanoparticles), there are many examples of "naturally" generated nanostructures - for example, extracellular vesicles (EVs), lipoproteins, and virus - that provide useful information about human physiology or clinical conditions. For example, the detection of tumor-related exosomes, a specific type of EVs, in circulating fluids has been contributing to the diagnosis of cancer in an early stage. However, scientists have struggled to find a simple, fast, and low-cost method to accurately detect/identify these nanoparticles, since the majority of them have diameters between 100 and 150 nm, thus being far below the diffraction limit. Methods: This study investigated if, by projecting the information provided from short-term portions of the back-scattered laser light signal collected by a polymeric lensed optical fiber tip dipped into a solution of synthetic nanoparticles into a lower features dimensional space, a discriminant function is able to correctly detect the presence of 100 nm synthetic nanoparticles in distilled water, in different concentration values. Results and discussion: This technique ensured an optimal performance (100% accuracy) in detecting nanoparticles for a concentration above or equal to 3.89 mu g/mL (8.74E+10 particles/mL), and a performance of 90% for concentrations below this value and higher than 1.22E-03 mu g/mL (2.74E+07 particles/mL), values that are compatible with human plasmatic levels of tumor-derived and other types of EVs, as well as lipoproteins currently used as potential biomarkers of cardiovascular diseases. Conclusion: The proposed technique is able to detect synthetic nanoparticles whose dimensions are similar to EVs and other "clinically" relevant nanostructures, and in concentrations equivalent to the majority of cell-derived, platelet-derived EVs and lipoproteins physiological levels. This study can, therefore, provide valuable insights towards the future development of a device for EVs and other biological nanoparticles detection with innovative characteristics.