We consider an online two-stage Bayesian group testing protocol. In the first stage, a Bayesian estimator is utilized to identify active items. In the second stage, a subset of the items are verified. The key challenge for Bayesian group testing that we address in this paper is imperfect knowledge of the prior distribution. We propose a new scheme based on gradient boosting where the estimator in the first stage utilizes convex combinations of the posteriors for a set of base estimators exploiting imperfect priors. We analyze the regret and show via a numerical study that our algorithm can outperform a baseline estimator with perfect knowledge of the prior distribution that does not exploit verification.

Federated learning (FL) is now recognized as a key framework for communication-efficient collaborative learning. Most theoretical and empirical studies, however, rely on the assumption that clients have access to pre-collected data sets, with limited investigation into scenarios where clients continuously collect data. In many real-world applications, particularly when data is generated by physical or biological processes, client data streams are often modeled by non-stationary Markov processes. Unlike standard i.i.d. sampling, the performance of FL with Markovian data streams remains poorly understood due to the statistical dependencies between client samples over time. In this paper, we investigate whether FL can still support collaborative learning with Markovian data streams. Specifically, we analyze the performance of Minibatch SGD, Local SGD, and a variant of Local SGD with momentum. We answer affirmatively under standard assumptions and smooth non-convex client objectives: the sample complexity is proportional to the inverse of the number of clients with a communication complexity comparable to the i.i.d. scenario. However, the sample complexity for Markovian data streams remains higher than for i.i.d. sampling.

Tailoring low resolution quantization to specific tasks is an important aspect of goal-oriented communications, with applications ranging from feedback for multiple-input multiple-output (MIMO) systems to compression of neural network weights. However, target distortion functions associated with a task may either be imperfectly known or intractable to compute, leading to uncertainty in the distortion function. In this letter, we propose new criteria based on risk measures to design 1-bit quantizers with uncertainty in the distortion function. While information-theoretic limits of compression with risk measure constraints have recently been established, quantizers based on risk measure criteria have not yet been developed. We introduce an algorithm for quantizer design with our risk measure criteria based on the cross-entropy method, and apply our approach to an environmental monitoring dataset.

A standard assumption in the design of ultra-reliable low-latency communication systems is that the duration between message arrivals is larger than the number of channel uses before the decoding deadline. Nevertheless, this assumption fails when messages arrive rapidly and reliability constraints require that the number of channel uses exceed the time between arrivals. In this paper, we consider a broadcast setting in which a transmitter wishes to send two different messages to two receivers over Gaussian channels. Messages have different arrival times and decoding deadlines such that their transmission windows overlap. For this setting, we propose a coding scheme that exploits Marton's coding strategy. We derive rigorous bounds on the achievable rate regions. Those bounds can be easily employed in point-to-point settings with one or multiple parallel channels. In the point-to-point setting with one or multiple parallel channels, the proposed achievability scheme is consistent with the normal approximation. In the broadcast setting, our scheme agrees with Marton's strategy for sufficiently large numbers of channel uses and shows significant performance improvements over standard approaches based on time sharing for transmission of short packets.

Inference often relies on compressed data due to communication, storage, or privacy constraints. In order to minimize degradation in the quality of inference, it is desirable to tailor compression schemes to the inference task. The compression scheme should therefore account for the statistic of the loss relevant for the task. While the expected loss is widely considered, in applications sensitive to large losses—such as in safe control and learning—alternative statistics are relevant. A key family of these alternative statistics are obtained via risk measures. In this paper, we characterize the increase in risk measure criteria for inference tasks as a function of the code size. Our characterization applies for general data statistics, loss functions, and number of samples. In the special case of i.i.d. data, we also establish asymptotics and a connection between our characterization for risk measure criteria and the rate-distortion function, which was previously only known for expected loss and excess distortion criteria.

AI-based sensing at wireless edge devices has the potential to significantly enhance Artificial Intelligence (AI) applications, particularly for vision and perception tasks such as in autonomous driving and environmental monitoring. AI systems rely both on efficient model learning and inference. In the inference phase, features extracted from sensing data are utilized for prediction tasks (e.g., classification or regression). In edge networks, sensors and model servers are often not co-located, which requires communication of features. As sensitive personal data can be reconstructed by an adversary, transformation of the features are required to reduce the risk of privacy violations. While differential privacy mechanisms provide a means of protecting finite datasets, protection of individual features has not been addressed. In this paper, we propose a novel framework for privacy-preserving AI-based sensing, where devices apply transformations of extracted features before transmission to a model server.

In the context of the Internet of Bio-Nano Things (IoBNT), nano-devices are envisioned to perform complex tasks collaboratively, i.e., by communicating with each other. One candidate for the implementation of such devices are engineered cells due to their inherent biocompatibility. However, because each engineered cell has only little computational capabilities, transmitter and receiver (RX) functionalities can afford only limited complexity. In this paper, we propose a simple, yet modular, architecture for a cellular RX that is capable of processing a stream of observed symbols using chemical reaction networks. Furthermore, we propose two specific detector implementations for the RX. The first detector is based on a machine learning model that is trained offline, i.e., before the cellular RX is deployed. The second detector utilizes pilot symbol-based training and is therefore able to continuously adapt to changing channel conditions online, i.e., after deployment. To coordinate the different chemical processing steps involved in symbol detection, the proposed cellular RX leverages an internal chemical timer. Furthermore, the RX is synchronized with the transmitter via external, i.e., extracellular, signals. Finally, the proposed architecture is validated using theoretical analysis and stochastic simulations. The presented results confirm the feasibility of both proposed implementations and reveal that the proposed online learning-based RX is able to perform reliable detection even in initially unknown or slowly changing channels. By its modular design and exclusively chemical implementation, the proposed RX contributes towards the realization of versatile and biocompatible nano-scale communication networks for IoBNT applications narrowing the existing implementation gap in cellular molecular communication (MC).

Grant-free random access (GFRA) is now a popular protocol for large-scale wireless multiple access systems in order to reduce control signaling. Resource allocation in GFRA can be viewed as a form of frame slotted ALOHA, where a ubiquitous design assumption is device homogeneity. In particular, the probability that a device seeks to transmit data is common to all devices. Recently, there has been an interest in designing frame slotted ALOHA algorithms for networks with heterogeneous activity probabilities. These works have established that the throughput can be significantly improved over the standard uniform allocation. However, the algorithms for optimizing the probability a device accesses each slot require perfect knowledge of the active devices within each frame. In practice, this assumption is limiting as device identification algorithms in GFRA rarely provide activity estimates with zero errors. In this paper, we develop a new algorithm based on stochastic gradient descent for optimizing slot allocation probabilities in the presence of activity estimation errors. Our algorithm exploits importance weighted bias mitigation for stochastic gradient estimates, which is shown to provably converge to a stationary point of the throughput optimization problem. In moderate size systems, our simulations show that the performance of our algorithm depends on the type of error distribution. We study symmetric bit flipping, asymmetric bit flipping and errors resulting from a generalized approximate message passing (GAMP) algorithm. In these scenarios, we observe gains up to 40%, 66%, and 19%, respectively.

Grant Free Random Access (GFRA) is a popular protocol in the Internet of Things (IoT) to reduce the control signaling. GFRA is a framed protocol where each frame is split into two parts: device identification; and data transmission part which can be viewed as a form of Frame Slotted ALOHA (FSA). A common assumption in FSA is device homogeneity; that is the probability that a device seeks to transmit data in a particular frame is common for all devices and independent of the other devices. Recent work has investigated the possibility of tuning the FSA protocol to the statistics of the network by changing the probability for a particular device to access a particular slot. However, power control with a successive interference cancellation (SIC) receiver has not yet been considered to further increase the performance of the tuned FSA protocols. In this paper, we propose algorithms to jointly optimize both the slot selection and the transmit power of the devices to minimize the outage of the devices in the network. We show via a simulation study that our algorithms can outperform baselines (including slotted ALOHA) in terms of expected number of devices transmitting without outage and in term of transmit power.

On 25 July 2023, a workshop entitled “Molecular Communication Approaches for Wetware Artificial Life” took place as a satellite event at the international conference ALIFE 2023 (The 2023 Conference on Artificial Life). In this report, we comment on the workshop by focusing on the main theme and the motivations that led us to develop this initiative. In particular, we highlight how recent progress in synthetic biology and in the study of molecular communication from an engineering perspective can be fruitfully joined to provide a powerful platform to develop frontier research lines in “wetware” Artificial Life. The talks presented at the workshop are briefly summarized. This report is, ultimately, an opportunity to promote an emerging field that calls for collaborative efforts of scholars from multiple disciplines, from chemistry to molecular biology, from communication engineering to nanotechnology, and up to those interested in more theoretical aspects about complex artificial systems that mimic natural ones.

To model bio-chemical reaction systems with diffusion one can either use stochastic, microscopic reaction-diffusion master equations or deterministic, macroscopic reaction-diffusion system. The connection between these two models is not only theoretically important but also plays an essential role in applications. This paper considers the macroscopic limits of the chemical reaction-diffusion master equation for first-order chemical reaction systems in highly heterogeneous environments. More precisely, the diffusion coefficients as well as the reaction rates are spatially inhomogeneous and the reaction rates may also be discontinuous. By carefully discretizing these heterogeneities within a reaction-diffusion master equation model, we show that in the limit we recover the macroscopic reaction-diffusion system with inhomogeneous diffusion and reaction rates.

The random access procedure is a bottleneck to the development of wireless networks supporting the use cases of massive machine-type communication and ultra reliable and low-latency communication. Such networks are densely and massively popupated and must meet stringent latency and reliability requirements. Due to these characteristics, grant-free random access is envisioned to alleviate the control overhead generated by the classical random access procedure. However, active user detection and channel estimation algorithms are required. Existing algorithms assume that the activity of each device is homogeneous and independent, which is not the case in many applications (e.g., due to sensors observing a common phenomenon). In order to address this problem, we introduce a new flexible model taking into account a group-heterogeneous activity, using the framework of copula theory. It is then leveraged by a hybrid generalized approximate message passing algorithm to solve the active user detection and channel estimation problem. Our numerical results show that the user detection and channel estimation are both improved with this new algorithm w.r.t. state-of-the-art Bayesian algorithms, with gains up to 10 times fewer detection errors and 10 dB less channel estimation error.

A challenge for real-time monitoring of biochemical processes, such as cells, is detection of biologically relevant molecules. This is due to the fact that spectroscopy methods for detection may perturb the cellular environment. One approach to overcome this problem is coupled microfluidic-spectroscopy, where a microfluidic output channel is introduced in order to observe biologically relevant molecules. This approach allows for non-passive spectroscopy methods, such as mass spectrometry, to identify the structure of molecules released by the cell. Due to the non-negligible length of the microfluidic channel, when a sequence of stimuli are applied to a cell it is not straightforward to determine which spectroscopy samples correspond to a given stimulus. In this paper, we propose a solution to this problem by taking a molecular communication (MC) perspective on the coupled microfluidic-spectroscopy system. In particular, assignment of samples to a stimulus is viewed as a synchronization problem. We develop two new algorithms for synchronization in this context and carry out a detailed theoretical and numerical study of their performance. Our results show improvements over maximum-likelihood synchronization algorithms in terms of detection performance when there are uncertainties in the composition of the microfluidic channel.

In grant-free random access, a key question is how devices should utilize resources without coordination. One standard solution to this problem are strategies where devices randomly select time-slots based on an optimized stochastic allocation rule. However, the optimization of this allocation rule requires accurate knowledge of which devices have been active in previous frames. As user identification algorithms are subject to errors, the expected throughput of the optimized allocation can be highly suboptimal. In this paper, we propose algorithms for optimization of device time-slot allocations that mitigate the impact of user identification errors. We show that when the activity distribution with and without errors is known, then our algorithm converges with probability one to a stationary point. When the activity distributions are not available, we introduce new theoretically-motivated heuristics which significantly improve the expected throughput over existing algorithms and approach the performance when errors are not present.

The training energy efficiency of deep neural networks became an extensively studied research topic in the last years. Some of the existing approaches seek to reduce the size of the architecture by either starting the training with a large network and pruning it, or by beginning with a seed architecture and then growing it. Instead of compressing the architecture, other approaches aim to reduce the number of training examples through data selection. While various approaches belonging to these two categories have been proposed, only a few works actually conduct energy measurements. Others merely mention potential gains in efficiency or rely on alternative evaluation metrics such as FLOPs. In this paper, we conduct a series of experiments both on a synthetic dataset and on image classification benchmarks in order to compare the impact of pruning, architecture growing and data selection on training energy consumption and prediction quality. Our results show that growing maintains a high prediction quality but brings limited energy gains when the size of the resulting architecture is large. Pruning can offer high gains, but also impacts accuracy, making it more suited for large models. Data selection provides energy gains correlated with the selectivity rate but causes an accuracy loss. We find that the effectiveness of every technique depends on its hyperparameters and on the architecture size.

In predictive maintenance an important problem is to optimize the quantity of information to be transmitted at the control center to guarantee reliable fault detection while limiting sensor power consumption. This problem relies directly on the sensor configurations (e.g., sampling rate, coding, quantization) and the fault detection algorithm. To address this question, we introduce a codesign framework and an algorithm for joint optimization of the sensor configurations and the accuracy of the fault detection classifier. In a use case based on a dataset consisting of multiple sensor measurements and heating power levels known as the Twin House Experiment, we show that our algorithm can find efficient tradeoffs between sensor power consumption and classifier accuracy.

In order to fully exploit the potential of molecular communication (MC) for intra-body communication, practically implementable cellular receivers are an important long-term goal. A variety of receiver architectures based on chemical reaction networks (CRNs) and gene-regulatory networks (GRNs) has been introduced in the literature, because cells use these concepts to perform computations in nature. However, practical feasibility is still limited by stochastic fluctuations of chemical reactions and long computation times in GRNs. Therefore, in this paper, we propose two receiver designs based on stochastic CRNs, i.e., CRNs that perform computations by exploiting the intrinsic fluctuations of chemical reactions with very low molecule counts. The first CRN builds on a recent result from chemistry that showed how Boltzmann machines (BMs), a commonly used machine learning model, can be implemented with CRNs. We show that BMs with optimal parameter values and their CRN implementations can act as maximum-a-posteriori (MAP) detectors. Furthermore, we show that BMs can be efficiently trained from simulation data to achieve close-to-MAP performance. While this approach yields a fixed CRN once deployed, our second approach based on a manually designed CRN can be trained with pilot symbols even within the cell and thus adapt to changing channel conditions. We extend the literature by showing that practical robust detectors can achieve close-to-MAP performance even without explicit channel knowledge.

Within many chemical and biological systems, both synthetic and natural, communication via chemical messengers is widely viewed as a key feature. Often known as molecular communication, such communication has been a concern in the fields of synthetic biologists, nanotechnologists, communications engineers, and philosophers of science. However, interactions between these fields are currently limited. Nevertheless, the fact that the same basic phenomenon is studied by all of these fields raises the question of whether there are unexploited interdisciplinary synergies. In this paper, we summarize the perspectives of each field on molecular communications, highlight potential synergies, discuss ongoing challenges to exploit these synergies, and present future perspectives for interdisciplinary efforts in this area.