The performance of machine learning algorithms is limited by the availability of training data. Transfer learning can alleviate this limitation by adapting models trained in data-rich domains to a data-sparse domain. In this work, we propose a method for data augmentation in a partially sampled datasparse target domain by transporting structural insights on data from a data-rich source domain. As an example, we show how this approach can improve the performance of a battery core temperature estimation based on electrochemical impedance spectroscopy (EIS) measurements, when only limited training data is available for new battery chemistries and form factors.

In-bed, non-contact vital signs monitoring enables patient monitoring without the use of intrusive sensors that interfere with patients’ sleep. Accelerometers incorporated into mattresses can be used to estimate insights related to vitals and sleep quality when someone lays on the mattress. A necessary first step to this is detecting whether someone is lying on the bed. In this signal processing grand challenge, we consider the problem of person-in-bed detection using only 3-axis accelerometer data obtained using the ADXL355 sensor. In particular, we focus on low-cost, low-latency detection algorithms through two challenge tracks, the first focused on detection over limited windows and the second on state tracking aimed to detect changes quickly.

Blockchains store transaction data in the form of a distributed ledger where each node in the network stores a current copy of the sequence of transactions as a hash chain. This requirement of storing the entire ledger incurs a high storage cost that grows undesirably large for high transaction rates and large networks. In this work we use secret key sharing, private key encryption, and distributed storage to design a coding scheme such that each node stores only a part of each transaction, thereby reducing the cold storage cost to a fraction of its original cost. In addition, the storage code ensures the security of the storage from active adversaries that may aim to corrupt prior transactions by altering copies of the ledger. We further employ a dynamic zone allocation algorithm that spreads the node allocation and data distribution across transactions. Under this coding scheme we show that we can also improve the integrity of the transaction data in the network over current schemes.

We study the problem of image registration in the finite-resolution regime and characterize the error probability of algorithms as a function of properties of the transformation and the image capture noise. Specifically, we define a channel-aware Feinstein decoder to obtain upper bounds on the minimum achievable error probability under finite resolution. We specifically focus on the higher-order terms and use Berry-Esseen type CLTs to obtain a stronger characterization of the achievability condition for the problem. Then, we derive a strong type-counting result to characterize the performance of the MMI decoder in terms of the maximum likelihood decoder, in a simplified setting of the problem. We then describe how this analysis, when related to the results from the channel-aware context provide stronger characterization of the finite-sample performance of universal image registration.

Consider a social learning problem in a parallel network, where N distributed agents make independent selfish binary decisions, and a central agent aggregates them together with a private signal to make a final decision. In particular, all agents have private beliefs for the true prior, based on which they perform binary hypothesis testing. We focus on the Bayes risk of the central agent, and counterintuitively find that a collection of agents with incorrect beliefs could outperform a set of agents with correct beliefs. We also consider many-agent asymptotics (i.e., N is large) when distributed agents all have identical beliefs, for which it is found that the central agent's decision is polarized and beliefs determine the limit value of the central agent's risk. Moreover, it is surprising that when all agents believe a certain prior-agnostic constant belief, it achieves globally optimal risk as N → ∞.

In previous work, we proposed a scalable multi-party verification scheme for expensive iterative computations on a Blockchain substrate by appropriate storage and endorsement of frames of iterates. In this work, we extend the framework to verify sets of complete computations with different unordered hyperparameters and develop frame ordering and compression algorithms to enable scalability in the system. We illustrate the efficacy of the proposed approach by verifying the OpenMalaria epidemiological simulation.

We introduce a sequential Bayesian binary hypothesis testing problem under social learning, termed selfish learning, where agents work to maximize their individual rewards. In particular, each agent receives a private signal and is aware of decisions made by earlier-acting agents. Beside inferring the underlying hypothesis, agents also decide whether to stop and declare, or pass the inference to the next agent. The employer rewards only correct responses and the reward per worker decreases with the number of employees used for decision making. We characterize decision regions of agents in the infinite and finite horizon. In particular, we show that the decision boundaries in the infinite horizon are the solutions to a Markov Decision Process with discounted costs, and can be solved using value iteration. In the finite horizon, we show that team performance is enhanced upon appropriate incen-tivization when compared to sequential social learning.

This work explores sequential Bayesian binary hypothesis testing in the social learning setup under expertise diversity. We consider a two-agent (say advisor-learner) sequential binary hypothesis test where the learner infers the hypothesis based on the decision of the advisor, a prior private signal, and individual belief. In addition, the agents have varying expertise, in terms of the noise variance in the private signal. Under such a setting, we first investigate the behavior of optimal agent beliefs and observe that the nature of optimal agents could be inverted depending on expertise levels. We also discuss suboptimality of the Prelec reweighting function under diverse expertise. Next, we consider an advisor selection problem wherein the belief of the learner is fixed and the advisor is to be chosen for a given prior. We characterize the decision region for choosing such an advisor and argue that a learner with beliefs varying from the true prior often ends up selecting a suboptimal advisor.

Blockchain uses the idea of storing transaction data in the form of a distributed ledger wherein each node in the network stores a current copy of the sequence of transactions (ledger) in the form of a hash chain. Storing the entire ledger incurs a high storage cost that grows undesirably large for high transaction rates and large networks. In this work we use secret key sharing, private key encryption, and distributed storage to design a coding scheme such that each node stores only a part of each transaction thereby reducing storage cost to a fraction of the original. When further using dynamic zone allocation, we show the coding scheme can also improve the data integrity.

We address the problem of downlink resource allocation in the presence of time-varying interference. We consider a scenario where users served by a base station face interference from a neighboring base station. We model the interference from the neighboring base station as an ON/OFF renewal process, that arises due to its idle and busy cycles. The users feedback their downlink signal to interference plus noise ratio (SINR) values to their base station, but these values are outdated. In this setting, we characterize how the resource allocation layer can optimally exploit the reported SINR values, which could be unreliable due to time-varying interference. In particular, we propose resource allocation policies in two well-known paradigms. First, we address the problem of α-fair scheduling, and propose a policy that ensures asymptotic convergence to the optimal α-fair throughput. Second, we propose a throughput optimal resource allocation policy, i.e., a policy that can stably support the largest possible set of traffic rates under the interference scenario considered. Estimating the outage probability from the outdated SINR values plays an important role in both scheduling paradigms, and we accomplish this using tool from renewal theory.

This paper defines and studies the problem of universal clustering using responses of crowd workers, without knowledge of worker reliability or task difficulty. We model stochastic worker response distributions by incorporating traits of memory for similar objects and traits of distance among differing objects. We are particularly interested in two limiting worker types - temporary and long-term workers, without and with memory respectively. We first define clustering algorithms for these limiting cases and then integrate them into an algorithm for the unified worker model. We prove asymptotic consistency of the algorithms and establish sufficient conditions on the sample complexity of the algorithm. Converse arguments establish necessary conditions on sample complexity, proving that the defined algorithms are asymptotically order-optimal in cost.

The problem of joint clustering and registration of images is studied in a universal setting. We define universal joint clustering and registration algorithms using multivariate information functionals. We first study the problem of registering two images using maximum mutual information and prove its asymptotic optimality. We then show the shortcomings of pairwise registration in multi-image registration, and design an asymptotically optimal algorithm based on multi-information. Finally, we define a novel multivariate information functional to perform joint clustering and registration of images, and prove consistency of the algorithm.