Accurate solar forecasting underpins effective renewable energy management. We present SolarCAST, a causally informed model predicting future global horizontal irradiance (GHI) at a target site using only historical GHI from site X and nearby stations S---unlike prior work that relies on sky-camera or satellite imagery requiring specialized hardware and heavy preprocessing. To deliver high accuracy with only public sensor data, SolarCAST models three classes of confounding factors behind X-S correlations using scalable neural components: (i) observable synchronous variables (e.g., time of day, station identity), handled via an embedding module; (ii) latent synchronous factors (e.g., regional weather patterns), captured by a spatio-temporal graph neural network; and (iii) time-lagged influences (e.g., cloud movement across stations), modeled with a gated transformer that learns temporal shifts. It outperforms leading time-series and multimodal baselines across diverse geographical conditions, and achieves a 25.9% error reduction over the top commercial forecaster, Solcast. SolarCAST offers a lightweight, practical, and generalizable solution for localized solar forecasting. Code available at https://github.com/YananNiu/SolarCAST

Accurate intraday solar irradiance forecasting is crucial for optimizing dispatch planning and electricity trading. For this purpose, we introduce a novel and effective approach that includes three distinguishing components from the literature: 1) the uncommon use of single-frame public camera imagery; 2) solar irradiance time series scaled with a proposed normalization step, which boosts performance; and 3) a lightweight multimodal model, called Solar Multimodal Transformer (SMT), that delivers accurate shortterm solar irradiance forecasting by combining images and scaled time series. Benchmarking against Solcast, a leading solar forecasting service provider, our model improved prediction accuracy by 25.95%. Our approach allows for easy adaptation to various camera specifications, offering broad applicability for real-world solar forecasting challenges.

Diffusion models generate new samples by progressively decreasing the noise from the initially provided random distribution. This inference procedure generally utilizes a trained neural network numerous times to obtain the final output, creating significant latency and energy consumption on digital electronic hardware such as GPUs. In this study, we demonstrate that the propagation of a light beam through a transparent medium can be programmed to implement a denoising diffusion model on image samples. This framework projects noisy image patterns through passive diffractive optical layers, which collectively only transmit the predicted noise term in the image. The optical transparent layers, which are trained with an online training approach, backpropagating the error to the analytical model of the system, are passive and kept the same across different steps of denoising. Hence this method enables high-speed image generation with minimal power consumption, benefiting from the bandwidth and energy efficiency of optical information processing.

In the early visual system, high dimensional natural stimuli are encoded into the trains of neuronal spikes that transmit the information to the brain to produce perception. However, is all the visual scene information required to explain the neuronal responses? In this work, we search for answers to this question by developing a joint model of the natural visual input and neuronal responses using the Information Bottleneck (IB) framework that can represent features of the input data into a few latent variables that play a role in the prediction of the outputs. The correlations between data samples acquired from published experiments on ex-vivo retinas are accounted for in the model by a Gaussian Process (GP) prior. The proposed IB-GP model performs competitively to the state-of-the-art feedforward convolutional networks in predicting spike responses to natural stimuli. Finally, the IB-GP model is used in a closed-loop iterative process to obtain reduced-complexity inputs that elicit responses as elicited by the original stimuli. We found three properties of the retina's IB-GP model. First, the reconstructed stimuli from the latent variables show robustness in spike prediction across models. Second, surprisingly the dynamics of the high-dimensional stimuli and RGCs' responses are very well represented in the embeddings of the IB-GP model. Third, the minimum stimuli consist of different patterns: Gabor-type locally high-frequency filters, on- and off-center Gaussians, or a mixture of both. Overall, this work demonstrates that the IB-GP model provides a principled approach for joint learning of the stimuli and retina codes, capturing dynamics of the stimuli-RGCs in the latent space which could help better understand the computation of the early visual system.

Learning Tomography (LT) is a nonlinear optimization algorithm for computationally imaging three-dimensional (3D) distribution of the refractive index in semi-transparent samples. Since the energy function in LT is generally non-convex, the solution it obtains is not guaranteed to be globally optimal. In this paper, we describe linear and nonlinear tomographic reconstruction methods and compare them numerically. We present a review of the LT and, in addition, we investigate the influence of the initialization and exemplify the effect of regularization on the convergence of the algorithm. In particular, we show that both are essential for high-quality imaging in strongly scattering scenarios.

For original article see ibid., vol.35, no.3, p.558-68, May 1989. The authors thank Jinwen Ma, Jianwei Wu, and Qiansheng Cheng for pointing out the overcount in the original proof.

We report a novel hyperspectral optical sensor capable of providing image information with four degrees of freedom (4D), ie, volumetric spatial information and spectral information simultaneously, in real time. The imaging principle is based on the diffraction properties of volume holographic optical elements designed as spatial-spectral filters. The 4D imager can be configured to image objects in spatial scales ranging from meters to micrometers. The imager also enables feature matching in all four dimensions, which improves the information-extraction capabilities from the image data. We report experimental results and theoretical estimates on the image quality attainable by the 4D imager.

WE examine the present state of holographic random access memory (HRAM) systems and address the primary challenges that face this technology, specifically size, speed, and cost. We show that a fast HRAM system can be implemented with a compact architecture by incorporating conjugate readout, a smart-pixel array, and a linear array of laser diodes. Preliminary experimental results support the feasibility of this architecture. Our analysis shows that in order for the HRAM to become competitive, the principal tasks will be to reduce spatial light modulator (SLM) and detector pixel sizes to I mu m, increase the output power of compact visible-wavelength lasers to several hundred milliwatts, and develop ways to raise the sensitivity of holographic media to the order of 1 cm/J.

The lower and upper bounds for the information capacity of two-layer feedforward neural networks with binary interconnections, integer thresholds for the hidden units, and zero threshold for the output unit is obtained through two steps. First, through a constructive approach based on statistical analysis, it is shown that a specifically constructed (N-2L-1) network with N input units, 2L hidden units, and one output unit is capable of implementing, with almost probability one, any dichotomy of O(W/ln W) random samples drawn from some continuous distributions, where W is the total number of weights of the network. This quantity is then used as a lower bound for the information capacity C of all (N-2L-1) networks with binary weights. Second, an upper bound is obtained and shown to be O(W) by a simple counting argument. Therefore, we have /spl Omega/(W/ln W)/spl les/C/spl les/O(W).

The finite sample performance of a nearest neighbor classifier is analyzed for a two-class pattern recognition problem. An exact integral expression is derived for the m-sample risk R/sub m/ given that a reference m-sample of labeled points is available to the classifier. The statistical setup assumes that the pattern classes arise in nature with fixed a priori probabilities and that points representing the classes are drawn from Euclidean n-space according to fixed class-conditional probability distributions. The sample is assumed to consist of m independently generated class-labeled points. For a family of smooth class-conditional distributions characterized by asymptotic expansions in general form, it is shown that the m-sample risk R/sub m/ has a complete asymptotic series expansion R/sub m//spl sim/R/sub /spl infin//+/spl Sigma//sub k=2//sup /spl infin//c/sub k/m/sup -k/n/ (m/spl rarr//spl infin/), where R/sub /spl infin// denotes the nearest neighbor risk in the infinite-sample limit and the coefficients c/sub k/ are distribution-dependent constants independent of the sample size m. The analysis thus provides further analytic validation of Bellman's curse of dimensionality. Numerical simulations corroborating the formal results are included, and extensions of the theory discussed. The analysis also contains a novel application of Laplace's asymptotic method of integration to a multidimensional integral where the integrand attains its maximum on a continuum of points.>

The finite sample performance of a nearest neighbor classifier is analyzed for a two-class pattern recognition problem. An exact integral expression is derived for the m-sample risk Rm given that a reference m-sample of labeled points, drawn independently from Euclidean n-space according to a fixed probability distri bution, is available to the classifier. For a family of smooth distributions, it is shown that the m-sample risk Rm has a complete asymptotic expansion ******, where ** denotes the nearest neighbor risk in the infinite sample limit. Explicit definitions of the expansion coefficents are given in terms of the underlying distribution. As the convergence rate of **** dramatically slows down as n increases, this analysis provides an analytic validation of Bellman's curse of dimensionality. Numerical simulations corroborating the formal results are included. The rates of convergence for less restrictive families of distributions are also discussed.

The authors investigate the computing capabilities of formal McCulloch-Pitts neurons when errors are permitted in decisions. They assume that m decisions are to be made on a randomly specified m set of points in n space and that an error tolerance of epsilon m decision errors is allowed, with 0>

We study how certain smoothness constraints, for example, piecewise continuity, can be generalized from a discrete set of analog-valued data, by modifying the error backpropagation, learning algorithm. Numerical simulations demonstrate that by imposing two heuristic objectives — (1) reducing the number of hidden units, and (2) minimizing the magnitudes of the weights in the network — during the learning process, one obtains a network with a response function that smoothly interpolates between the training data.

A model of associate memory incorporating global linearity and pointwise nonlinearities in a state space of n-dimensional binary vectors is considered. Attention is focused on the ability to store a prescribed set of state vectors as attractors within the model. Within the framework of such associative nets, a specific strategy for information storage that utilizes the spectrum of a linear operator is considered in some detail. Comparisons are made between this spectral strategy and a prior scheme that utilizes the sum of Kronecker outer products of the prescribed set of state vectors, which are to function nominally as memories. The storage capacity of the spectral strategy is linear in n (the dimension of the state space under consideration), whereas an asymptotic result of n/4 log n holds for the storage capacity of the outer product scheme. Computer-simulated results show that the spectral strategy stores information more efficiently. The preprocessing costs incurred in the two algorithms are estimated, and recursive strategies are developed for their computation.>

The role of moments in image normalization and invariant pattern recognition is addressed. The classical idea of the principal axes is analyzed and extended to a more general definition. The relationship between moment-based normalization, moment invariants, and circular harmonics is established. Invariance properties of moments, as opposed to their recognition properties, are identified using a new class of normalization procedures. The application of moment-based normalization in pattern recognition is demonstrated by experiment.

Moment invariants are evaluated as a feature space for pattern recognition in terms of discrimination power and noise tolerance. The notion of complex moments is introduced as a simple and straightforward way to derive moment invariants. Through this relation, properties of complex moments are used to characterize moment invariants. Aspects of information loss, suppression, and redundancy encountered in moment invariants are investigated and significant results are derived. The behavior of moment invariants in the presence of additive noise is also described.