Spiking Neural Networks (SNNs) are biologically-inspired deep neural networks that efficiently extract temporal information while offering promising gains in terms of energy efficiency and latency when deployed on neuromorphic devices. SNN parameters are sensitive to temporal resolution, leading to significant performance drops when the temporal resolution of target data during deployment is not the same as that of the source data used for training, especially when fine-tuning with the target data is not possible during deployment. To address this challenge, we propose three novel domain adaptation methods for adapting neuron parameters to account for the change in time resolution without re-training on target time resolution. The proposed methods are based on a mapping between neuron dynamics in SNNs and State Space Models (SSMs) and are applicable to general neuron models. We evaluate the proposed methods under spatio-temporal data tasks, namely the audio keyword spotting datasets SHD and MSWC, and the neuromorphic image NMINST dataset. Our methods provide an alternative to-and in most cases significantly outperform-the existing reference method that consists of scaling only the time constant. Notably, when the temporal resolution of the target data is double that of the source data, applying one of our proposed methods instead of the benchmark achieves classification accuracy of 89.5 % instead of 53.0 % on SHD, 93.6 % instead of 38.8 % on MSWC and 98.5 % instead of 97.2 % on NMNIST. Moreover, our results show that high accuracy on high temporal resolution data can be obtained by time-efficient training on lower temporal resolution data.

Imitation learning is an effective approach for training game-playing agents and, consequently, for efficient game production. However, generalization-the ability to perform well in related but unseen scenarios-is an essential requirement that remains an unsolved challenge for game AI. Generalization is difficult for imitation learning agents because it requires the algorithm to take meaningful actions outside of the training distribution. In this paper we propose a solution to this challenge. Inspired by the success of data augmentation in supervised learning, we augment the training data so the distribution of states and actions in the dataset better represents the real state-action distribution. This study evaluates methods for combining and applying data augmentations to observations, to improve generalization of imitation learning agents. It also provides a performance benchmark of these augmentations across several 3D environments. These results demonstrate that data augmentation is a promising framework for improving generalization in imitation learning agents.

We compute precise asymptotic expressions for the learning curves of least squares random feature (RF) models with either a separable strongly convex regularization or the $\ell_1$ regularization. We propose a novel multi-level application of the convex Gaussian min max theorem (CGMT) to overcome the traditional difficulty of finding computable expressions for random features models with correlated data. Our result takes the form of a computable 4-dimensional scalar optimization. In contrast to previous results, our approach does not require solving an often intractable proximal operator, which scales with the number of model parameters. Furthermore, we extend the universality results for the training and generalization errors for RF models to $\ell_1$ regularization. In particular, we demonstrate that under mild conditions, random feature models with elastic net or $\ell_1$ regularization are asymptotically equivalent to a surrogate Gaussian model with the same first and second moments. We numerically demonstrate the predictive capacity of our results, and show experimentally that the predicted test error is accurate even in the non-asymptotic regime.

Active inference (ActInf) is an emerging theory that explains perception and action in biological agents in terms of minimizing a free energy bound on Bayesian surprise. Goal-directed behavior is elicited by introducing prior beliefs on the underlying generative model. In contrast to prior beliefs, which constrain all realizations of a random variable, we propose an alternative approach through chance constraints, which allow for a (typically small) probability of constraint violation, and demonstrate how such constraints can be used as intrinsic drivers for goal-directed behavior in ActInf. We illustrate how chance-constrained ActInf weights all imposed (prior) constraints on the generative model, allowing, for example, for a trade-off between robust control and empirical chance constraint violation. Second, we interpret the proposed solution within a message passing framework. Interestingly, the message passing interpretation is not only relevant to the context of ActInf, but also provides a general-purpose approach that can account for chance constraints on graphical models. The chance constraint message updates can then be readily combined with other prederived message update rules without the need for custom derivations. The proposed chance-constrained message passing framework thus accelerates the search for workable models in general and can be used to complement message-passing formulations on generative neural models.

Random non-linear Fourier features have recently shown remarkable performance in a wide-range of regression and classification applications. Motivated by this success, this article focuses on a sparse non-linear Fourier feature (NFF) model. We provide a characterization of the sufficient number of data points that guarantee perfect recovery of the unknown parameters with high-probability. In particular, we show how the sufficient number of data points depends on the kernel matrix associated with the probability distribution function of the input data. We compare our results with the recoverability bounds for the bounded orthonormal systems and provide examples that illustrate sparse recovery under the NFF model.

This paper investigates the impact of the communication frequency on the remote control of automated vehicles. In particular, we consider a remote controller, which receives vehicles' state information and issues control commands based on a model predictive control (MPC) framework, to steer the vehicles to reach their respective target position intervals at given specific times. We present a framework where both state information (from the vehicles to the controller) and control actions (from the controller to the vehicles) are communicated through a wireless network. Due to limited communication resources and possible channel impairments, information is not necessarily always provided to the destination (either the controller or the vehicles). Herein, we particularly focus on the communications to the controller and investigate the effect of frequency and last instant of communication. Our results quantify the impact of these factors on the system performance, and subsequently, underline the need for an efficient resource allocation scheme.

The lifetime of a wireless sensor network (WSN) determines how long the network can be used to monitor the area of interest. Hence, it is one of the most important performance metrics for WSN. The approaches used to prolong the lifetime can be briefly divided into two categories: reducing the energy consumption, such as designing an efficient routing, and providing extra energy, such as using wireless energy transfer (WET) to charge the nodes. Contrary to the previous line of work where only one of those two aspects is considered, we investigate these two together. In particular, we consider a scenario where dedicated wireless chargers transfer energy wirelessly to sensors. The overall goal is to maximize the minimum sampling rate of the nodes while keeping the energy consumption of each node smaller than the energy it receives. This is done by properly designing the routing of the sensors and the WET strategy of the chargers. Although such a joint routing and energy beamforming problem is non-convex, we show that it can be transformed into a semi-definite optimization problem (SDP). We then prove that the strong duality of the SDP problem holds, and hence, the optimal solution of the SDP problem is attained. Accordingly, the optimal solution for the original problem is achieved by a simple transformation. We also propose a lowcomplexity approach based on pre-determined beamforming directions. Moreover, based on the alternating direction method of multipliers, the distributed implementations of the proposed approaches are studied. The simulation results illustrate the significant performance improvement achieved by the proposed methods. In particular, the proposed energy beamforming scheme significantly outperforms the schemes where one does not use energy beamforming, or one does not use optimized routing. A thorough investigation of the effect of system parameters, including the number of antennas, the number of nodes, and the number of chargers, on the system performance is provided. The promising convergence behavior of the proposed distributed approaches is illustrated.

Remote estimation with an energy harvesting sensor with a limited data and energy buffer is considered. The sensor node observes an unknown temporally correlated field and communicates its observations to a remote fusion center using the energy it harvested. The fusion center employs linear minimum mean-square error estimation to reconstruct the unknown field. We provide performance guarantees for the estimation error under a block transmission scheme, where at each transmission block, data and energy buffers are completely emptied. Our bounds provide insights into how statistical properties of the energy harvesting process and buffer sizes may affect the estimation error. In particular, these bounds suggest insensitivity of the performance to buffer sizes for signals with low degree of freedom and suggest performance improvements with increasing buffer sizes for signals with relatively higher degree of freedom. Depending only on the mean, variance, and finite support of the energy arrival process, these results provide insights for the energy and data buffer sizes for deployment in future energy harvesting wireless sensing systems.

Wireless sensor networks (WSNs) consist of energy limited sensor nodes, which limits the network lifetime. Such a lifetime can be prolonged by employing the emerging technology of wireless energy transfer (WET). In WET systems, the sensor nodes can harvest wireless energy from wireless charger, which can use energy beamforming to improve the efficiency. In this paper, a scenario where dedicated wireless chargers with multiple antennas use energy beamforming to charge sensor nodes is considered. The energy beamforming is coupled with the energy consumption of sensor nodes in terms of data routing, which is one novelty of the paper. The energy beamforming and the data routing are jointly optimized by a non-convex optimization problem. This problem is transformed into a semidefinite optimization problem, for which strong duality is proved, and thus the optimal solution exists. It is shown that the optimal solution of the semi-definite programming problem allows to derive the optimal solution of the original problem. The analytical and numerical results show that optimal energy beamforming gives two times better monitoring performance than that of WET without using energy beamforming.

We consider the remote estimation of a time-correlated signal using an energy harvesting (EH) sensor. The sensor observes the unknown signal and communicates its observations to a remote fusion center using an amplify-and-forward strategy. We consider the design of optimal power allocation strategies in order to minimize the mean-square error at the fusion center. Contrary to the traditional approaches, the degree of correlation between the signal values constitutes an important aspect of our formulation. We provide the optimal power allocation strategies for a number of illustrative scenarios. We show that the most majorized power allocation strategy, i.e., the power allocation as balanced as possible, is optimal for the cases of circularly wide-sense stationary (c.w.s.s.) signals with a static correlation coefficient and sampled low-pass c.w.s.s. signals for a static channel. We show that the optimal strategy can be characterized as a water-filling type solution for sampled low-pass c.w.s.s. signals for a fading channel. Motivated by the high-complexity of the numerical solution of the optimization problem, we propose low-complexity policies for the general scenario. Numerical evaluations illustrate the close performance of these low-complexity policies to that of the optimal policies and demonstrate the effect of the EH constraints and the degree of freedom of the signal.

Remote estimation with an energy harvesting sensor with a limited data buffer is considered. The sensor node observes an unknown correlated circularly wide-sense stationary (c.w.s.s.) Gaussian field and communicates its observations to a remote fusion center using the energy it harvested. The fusion center employs minimum mean-square error (MMSE) estimation to reconstruct the unknown field. The distortion minimization problem under the online scheme, where the sensor has only access to the statistical information for the future energy packets is considered. We provide performance bounds on the achievable distortion under a slotted block transmission scheme, where at each transmission time slot, the data and the energy buffer is completely emptied. Our bounds provide insight to the trade-off between the buffer sizes and the achievable distortion. These trade-offs illustrate the insensitivity of the performance to the buffer size for signals with low degree of freedom and suggest performance improvements with increasing buffer size for signals with relatively higher degree of freedom.

Simultaneous wireless information and power transfer (SWIPT) offers an attractive alternative to the traditional battery limited or grid dependent communication system design. In this paper, we investigate robust precoder designs for systems with SWIPT capabilities under a stochastic Rician fading framework. Under a multiple-input multiple-output (MIMO) channel model, we formulate the problem of minimizing the average mean-square error at the information receiver (IR) while keeping the average energy harvested at the energy receiver (ER) above given levels. We consider two different strategies that can be adopted by the IR: i) simple estimation filters based solely on the channel mean, ii) robust estimation filters aiming to minimize the average mean-square error. Both of these scenarios lead to non-convex formulations. For the first scenario, we propose a convex relaxation that is tight. For the second scenario, we propose an alternating minimization method that provides precoder designs even in the scenarios where the number of transmit antennas is larger than the length of data vector. Our numerical results show that the proposed designs provide significant performance gains especially when the scattering component of the channel is strong.

We focus on linear precoding strategies as a physical layer technique for providing security in Gaussian interference channels. We consider an artificial noise aided scheme where transmitters may broadcast noise in addition to data in order to confuse eavesdroppers. We formulate the problem of minimizing the total mean-square error at the legitimate receivers while keeping the error values at the eavesdroppers above target levels. This set-up leads to a non-convex problem formulation. Hence, we propose a coordinate block descent technique based on a tight semi-definite relaxation and design linear precoders as well as spatial distribution of the artificial noise. Our results illustrate that artificial noise can provide significant performance gains especially when the secrecy levels required at the eavesdroppers are demanding.

Communication strategies that utilize wireless media for simultaneous information and power transfer offer a promising perspective for efficient usage of energy resources. With this motivation, we focus on the design of optimal linear precoders for interference channels utilizing such strategies. We formulate the problem of minimizing the total minimum mean-square error while keeping the energy harvested at the energy receivers above given levels. Our framework leads to a non-convex problem formulation. For point-to-point multiple-input multiple-output channels, we provide a characterization of the optimal solutions under a constraint on the number of transmit antennas. For the general interference scenario, we propose two numerical approaches, one for the single antenna information receivers case, and the other for the general case. We also investigate a hybrid signalling scheme, where the transmitter sends a superposition of two signals: a deterministic signal optimized for energy transfer and an information carrying signal optimized for information and energy transfer. It is illustrated that if hybrid signalling is not incorporated into the transmission scheme, interference can be detrimental to the system performance when the number of antennas at the receivers is low.

We present lower bounds on the error probability of turbo codes under maximum likelihood (ML) decoding. We focus on additive white Gaussian noise (AWGN) channels, and consider both ensembles of codes with uniform interleaving and specific turbo codes with fixed interleavers. To calculate the lower bounds, instead of using the traditional approach that only makes use of the distance spectrum, we propose to utilize the exact second order distance spectrum. This approach together with a proper restriction of the error events results in promising lower bounds.

We focus on trellis-based joint code design for two-user Gaussian multiple-access channel (MAC) in the short block length regime. We propose a design methodology, provide specific code designs and report numerical performance results. We compare the performance of the jointly designed codes with the performance of the codes designed for point-to-point (P2P) channels including optimum (in terms of minimum distance) convolutional codes. Our results show that the proposed codes achieve superior performance compared to these alternatives especially in the high signal-to-noise (SNR) regime in equal power scenarios.

We consider the transmission of a Gaussian vector source over a multidimensional Gaussian channel where a random or a fixed subset of the channel outputs are erased. Within the setup where the only encoding operation allowed is a linear unitary transformation on the source, we investigate the minimum mean-square error (MMSE) performance, both in average, and also in terms of guarantees that hold with high probability as a function of the system parameters. Under the performance criterion of average MMSE, necessary conditions that should be satisfied by the optimal unitary encoders are established and explicit solutions for a class of settings are presented. For random sampling of signals that have a low number of degrees of freedom, we present MMSE bounds that hold with high probability. Our results illustrate how the spread of the eigenvalue distribution and the unitary transformation contribute to these performance guarantees. The performance of the discrete Fourier transform (DFT) is also investigated. As a benchmark, we investigate the equidistant sampling of circularly wide-sense stationary signals, and present the explicit error expression that quantifies the effects of the sampling rate and the eigenvalue distribution of the covariance matrix of the signal. These findings may be useful in understanding the geometric dependence of signal uncertainty in a stochastic process. In particular, unlike information theoretic measures such as entropy, we highlight the basis dependence of uncertainty in a signal with another perspective. The unitary encoding space restriction exhibits the most and least favorable signal bases for estimation.

In this paper, we study the effect of limited amplitude resolution (pixel depth) in super-resolution problem. The problem we address differs from the standard super-resolution problem in that amplitude resolution is considered as important as spatial resolution. We study the trade-off between the pixel depth and spatial resolution of low resolution (LR) images in order to obtain the best visual quality in the reconstructed high resolution (HR) image. The proposed framework reveals great flexibility in terms of pixel depth and number of LR images in super-resolution problem, and demonstrates that it is possible to obtain target visual qualities with different measurement scenarios including images with different amplitude and spatial resolutions.

We give a precise mathematical formulation of some measurement problems arising in optics, which is also applicable in a wide variety of other contexts. In essence the measurement problem is an estimation problem in which data collected by a number of noisy measurement probes are combined to reconstruct an unknown realization of a random process f(x) indexed by a spatial variable x epsi Rkfor some k ges 1. We wish to optimally choose and position the probes given the statistical characterization of the process f(x) and of the measurement noise processes. We use a model in which we define a cost function for measurement probes depending on their resolving power. The estimation problem is then set up as an optimization problem in which we wish to minimize the mean-square estimation error summed over the entire domain of f subject to a total cost constraint for the probes. The decision variables are the number of probes, their positions and qualities. We are unable to offer a solution to this problem in such generality; however, for the metrical problem in which the number and locations of the probes are fixed, we give complete solutions for some special cases and an efficient numerical algorithm for computing the best trade-off between measurement cost and mean-square estimation error. A novel aspect of our formulation is its close connection with information theory; as we argue in the paper, the mutual information function is the natural cost function for a measurement device. The use of information as a cost measure for noisy measurements opens up several direct analogies between the measurement problem and classical problems of information theory, which are pointed out in the paper.