Search

Fundamental limits on the error probabilities of a family of decentralized detection algorithms (eg., the social learning rule proposed by Lalitha et al. over directed graphs are investigated. In decentralized detection, a network of nodes locally exchanging information about the samples they observe with their neighbors to collectively infer the underlying unknown hypothesis. Each node in the network weighs the messages received from its neighbors to form its private belief and only requires knowledge of the data generating distribution of its observation. In this work, it is first shown that while the original social learning rule of Lalitha et al. achieves asymptotically vanishing error probabilities as the number of samples tends to infinity, it suffers a gap in the achievable error exponent compared to the centralized case. The gap is due to the network imbalance caused by the local weights that each node chooses to weigh the messages received from its neighbors. To close this gap, a modified learning rule is proposed and shown to achieve error exponents as large as those in the centralized setup. This implies that there is essentially no first-order penalty caused by decentralization in the exponentially decaying rate of error probabilities., Comment: Submitted to the IEEE Transactions on Information Theory

Artificial intelligence (AI) has revolutionized decision-making processes and systems throughout society and, in particular, has emerged as a significant technology in high-impact scenarios of national interest. Yet, despite AI's impressive predictive capabilities in controlled settings, it still suffers from a range of practical setbacks preventing its widespread use in various critical scenarios. In particular, it is generally unclear if a given AI system's predictions can be trusted by decision-makers in downstream applications. To address the need for more transparent, robust, and trustworthy AI systems, a suite of tools has been developed to quantify the uncertainty of AI predictions and, more generally, enable AI to "self-assess" the reliability of its predictions. In this manuscript, we categorize methods for AI self-assessment along several key dimensions and provide guidelines for selecting and designing the appropriate method for a practitioner's needs. In particular, we focus on uncertainty estimation techniques that consider the impact of self-assessment on the choices made by downstream decision-makers and on the resulting costs and benefits of decision outcomes. To demonstrate the utility of our methodology for self-assessment design, we illustrate its use for two realistic national-interest scenarios. This manuscript is a practical guide for machine learning engineers and AI system users to select the ideal self-assessment techniques for each problem.

Reliability of sequential hypothesis testing can be greatly improved when decision maker is given the freedom to adaptively take an action that determines the distribution of the current collected sample. Such advantage of sampling adaptivity has been realized since Chernoff's seminal paper in 1959 [1]. While a large body of works have explored and investigated the gain of adaptivity, in the general multiple-hypothesis setting, the fundamental limits of individual error probabilities have not been fully understood. In particular, in the asymptotic regime as the expected stopping time tends to infinity, the error exponents are only characterized in specific cases, such as that of the total error probability. In this paper, we consider a general setup of active sequential multiple-hypothesis testing where at each time slot, a temporally varying subset of data sources (out of a known set) emerges from which the decision maker can select to collect samples, subject to a family of expected selection budget constraints. The selection of sources, understood as the "action" at each time slot, is constrained in a predefined action space. At the end of each time slot, the decision maker either decides to make the inference on the $M$ hypotheses, or continues to observe the data sources for the next time slot. The optimal tradeoffs among $M(M-1)$ types of error exponents are characterized. A companion asymptotically optimal test that strikes the balance between exploration and exploitation is proposed to achieve any target error exponents within the region. To the best of our knowledge, this is the first time in the literature to identify such tradeoffs among error exponents, and it uncovers the tension among different action taking policies even in the basic setting of Chernoff [1]., Comment: Submitted to the IEEE Transactions on Information Theory

In the binary hypothesis testing problem, it is well known that sequentiality in taking samples eradicates the trade-off between two error exponents, yet implementing the optimal test requires the knowledge of the underlying distributions, say $P_0$ and $P_1$. In the scenario where the knowledge of distributions is replaced by empirically observed statistics from the respective distributions, the gain of sequentiality is less understood when subject to universality constraints over all possible $P_0,P_1$. In this work, the gap is mended by a unified study on sequentiality in the universal binary classification problem, where the universality constraints are set on the expected stopping time as well as the type-I error exponent. The type-I error exponent is required to achieve a pre-set distribution-dependent constraint $\lambda(P_0,P_1)$ for all $P_0,P_1$. Under the proposed framework, different sequential setups are investigated so that fair comparisons can be made with the fixed-length counterpart. By viewing these sequential classification problems as special cases of a general sequential composite hypothesis testing problem, the optimal type-II error exponents are characterized. Specifically, in the general sequential composite hypothesis testing problem subject to universality constraints, upper and lower bounds on the type-II error exponent are proved, and a sufficient condition for which the bounds coincide is given. The results for sequential classification problems are then obtained accordingly. With the characterization of the optimal error exponents, the benefit of sequentiality is shown both analytically and numerically by comparing the sequential and the fixed-length cases in representative examples of type-I exponent constraint $\lambda$., Comment: Submitted to the IEEE Transactions on Information Theory

Despite the progress in deep learning networks, efficient learning at the edge (enabling adaptable, low-complexity machine learning solutions) remains a critical need for defense and commercial applications. We envision a pipeline to utilize large neuroimaging datasets, including maps of the brain which capture neuron and synapse connectivity, to improve machine learning approaches. We have pursued different approaches within this pipeline structure. First, as a demonstration of data-driven discovery, the team has developed a technique for discovery of repeated subcircuits, or motifs. These were incorporated into a neural architecture search approach to evolve network architectures. Second, we have conducted analysis of the heading direction circuit in the fruit fly, which performs fusion of visual and angular velocity features, to explore augmenting existing computational models with new insight. Our team discovered a novel pattern of connectivity, implemented a new model, and demonstrated sensor fusion on a robotic platform. Third, the team analyzed circuitry for memory formation in the fruit fly connectome, enabling the design of a novel generative replay approach. Finally, the team has begun analysis of connectivity in mammalian cortex to explore potential improvements to transformer networks. These constraints increased network robustness on the most challenging examples in the CIFAR-10-C computer vision robustness benchmark task, while reducing learnable attention parameters by over an order of magnitude. Taken together, these results demonstrate multiple potential approaches to utilize insight from neural systems for developing robust and efficient machine learning techniques., Comment: 11 pages, 4 figures

In the "Beyond Moore's Law" era, with increasing edge intelligence, domain-specific computing embracing unconventional approaches will become increasingly prevalent. At the same time, adopting a variety of nanotechnologies will offer benefits in energy cost, computational speed, reduced footprint, cyber resilience, and processing power. The time is ripe for a roadmap for unconventional computing with nanotechnologies to guide future research, and this collection aims to fill that need. The authors provide a comprehensive roadmap for neuromorphic computing using electron spins, memristive devices, two-dimensional nanomaterials, nanomagnets, and various dynamical systems. They also address other paradigms such as Ising machines, Bayesian inference engines, probabilistic computing with p-bits, processing in memory, quantum memories and algorithms, computing with skyrmions and spin waves, and brain-inspired computing for incremental learning and problem-solving in severely resource-constrained environments. These approaches have advantages over traditional Boolean computing based on von Neumann architecture. As the computational requirements for artificial intelligence grow 50 times faster than Moore's Law for electronics, more unconventional approaches to computing and signal processing will appear on the horizon, and this roadmap will help identify future needs and challenges. In a very fertile field, experts in the field aim to present some of the dominant and most promising technologies for unconventional computing that will be around for some time to come. Within a holistic approach, the goal is to provide pathways for solidifying the field and guiding future impactful discoveries., Comment: 80 pages accepted in Nano Futures

Standard deep reinforcement learning (DRL) aims to maximize expected reward, considering collected experiences equally in formulating a policy. This differs from human decision-making, where gains and losses are valued differently and outlying outcomes are given increased consideration. It also fails to capitalize on opportunities to improve safety and/or performance through the incorporation of distributional context. Several approaches to distributional DRL have been investigated, with one popular strategy being to evaluate the projected distribution of returns for possible actions. We propose a more direct approach whereby risk-sensitive objectives, specified in terms of the cumulative distribution function (CDF) of the distribution of full-episode rewards, are optimized. This approach allows for outcomes to be weighed based on relative quality, can be used for both continuous and discrete action spaces, and may naturally be applied in both constrained and unconstrained settings. We show how to compute an asymptotically consistent estimate of the policy gradient for a broad class of risk-sensitive objectives via sampling, subsequently incorporating variance reduction and regularization measures to facilitate effective on-policy learning. We then demonstrate that the use of moderately "pessimistic" risk profiles, which emphasize scenarios where the agent performs poorly, leads to enhanced exploration and a continual focus on addressing deficiencies. We test the approach using different risk profiles in six OpenAI Safety Gym environments, comparing to state of the art on-policy methods. Without cost constraints, we find that pessimistic risk profiles can be used to reduce cost while improving total reward accumulation. With cost constraints, they are seen to provide higher positive rewards than risk-neutral approaches at the prescribed allowable cost., Comment: 16 pages, 13 figures. AAAI 2023 (Special Track on Safe and Robust AI)

We study the content delivery problem between a transmitter and two receivers through erasure links, when each receiver has access to some random side-information about the files requested by the other user. The random side-information is cached at the receiver via the decentralized content placement. The distributed nature of receiving terminals may also make the erasure state of two links and indexes of the cached bits not perfectly known at the transmitter. We thus investigate the capacity gain due to various levels of availability of channel state and cache index information at the transmitter. More precisely, we cover a wide range of settings from global delayed channel state knowledge and a non-blind transmitter (i.e. one that knows the exact cache index information at each receiver) all the way to no channel state information and a blind transmitter (i.e. one that only statistically knows cache index information at the receivers). We derive new inner and outer bounds for the problem under various settings and provide the conditions under which the two match and the capacity region is characterized. Surprisingly, for some interesting cases the capacity regions are the same even with single-user channel state or single-user cache index information at the transmitter.

We investigate the interaction between a moving detector and a quantum field, especially about how the trajectory of the detector would affect the vacuum fluctuations when the detector is moves in a quantum field (the Unruh effect). We focus on two moving detectors system for future application in quantum teleportation. We find that the rajectory of a uniformly accelerated detector in Rindler space cannot be extended to a trajectory in which a detector moves at constant velocity. Based on our previous work, we redo the calculations and find that a term is missing from the past calculations, and we also find that there are some restrictions on the values for the parameters in the solutions. In addition, without inclusion of the missing term, the variance from the quantum field for the inertial detector will be zero and is unlikely in such a system. When all these points are combined, there is a difference in the two-point correlation function between the inertial detector and the accelerated detector in the early-time region. The influence of proper acceleration can be seen in the two-point correlation functions. This might play a role in the quantum teleportation process and be worth studying thoroughly., Comment: 29 pages, 11 figures

Accurate state and uncertainty estimation is imperative for mobile robots and self driving vehicles to achieve safe navigation in pedestrian rich environments. A critical component of state and uncertainty estimation for robot navigation is to perform robustly under out-of-distribution noise. Traditional methods of state estimation decouple perception and state estimation making it difficult to operate on noisy, high dimensional data. Here, we describe an approach that combines the expressiveness of deep neural networks with principled approaches to uncertainty estimation found in recursive filters. We particularly focus on techniques that provide better robustness to out-of-distribution noise and demonstrate applicability of our approach on two scenarios: a simple noisy pendulum state estimation problem and real world pedestrian localization using the nuScenes dataset. We show that our approach improves state and uncertainty estimation compared to baselines while achieving approximately 3x improvement in computational efficiency.

In this paper, combinatorial quantitative group testing (QGT) with noisy measurements is studied. The goal of QGT is to detect defective items from a data set of size $n$ with counting measurements, each of which counts the number of defects in a selected pool of items. While most literatures consider either probabilistic QGT with random noise or combinatorial QGT with noiseless measurements, our focus is on the combinatorial QGT with noisy measurements that might be adversarially perturbed by additive bounded noises. Since perfect detection is impossible, a partial detection criterion is adopted. With the adversarial noise being bounded by $d_n = \Theta(n^\delta)$ and the detection criterion being to ensure no more than $k_n = \Theta(n^\kappa)$ errors can be made, our goal is to characterize the fundamental limit on the number of measurement, termed \emph{pooling complexity}, as well as provide explicit construction of measurement plans with optimal pooling complexity and efficient decoding algorithms. We first show that the fundamental limit is $\frac{1}{1-2\delta}\frac{n}{\log n}$ to within a constant factor not depending on $(n,\kappa,\delta)$ for the non-adaptive setting when $0<2\delta\leq \kappa <1$, sharpening the previous result by Chen and Wang [2]. We also provide an explicit construction of a non-adaptive deterministic measurement plan with $\frac{1}{1-2\delta}\frac{n}{\log_{2} n}$ pooling complexity up to a constant factor, matching the fundamental limit, with decoding complexity being $o(n^{1+\rho})$ for all $\rho > 0$, nearly linear in $n$, the size of the data set.