The emerging paradigm of federated learning strives to enable collaborative training machine models on the network edge without centrally aggregating raw data and hence, improving privacy. This sharply deviates from traditional necessitates design algorithms robust various sources heterogeneity. Specifically, statistical heterogeneity across user devices can severely degrade performance standard averaging for applications like personalization with deep learning. paper pro-posesFedPer, a base...

We examine the problem of utilizing an autonomous underwater vehicle (AUV) to collect data from sensor network. The sensors in network are equipped with acoustic modems that provide noisy, range-limited communication. AUV must plan a path maximizes information collected while minimizing travel time or fuel expenditure. propose planning methods extend algorithms for variants Traveling Salesperson Problem (TSP). While executing path, can improve performance by communicating multiple nodes at...

Cross language-image modality retrieval in E-commerce is a fundamental problem for product search, recommendation, and marketing services. Extensive efforts have been made to conquer the cross-modal general domain. When it comes E-commerce, com-mon practice adopt pretrained model finetune on data. Despite its simplicity, performance sub-optimal due overlooking uniqueness of multimodal A few recent [10], [72] shown significant improvements over generic methods with customized designs handling...

Human activity prediction is an interesting problem with a wide variety of applications like intelligent virtual assistants, contextual marketing, etc. One formulation this jointly predicting human activities (viz. eating, commuting, etc.) associated durations. Herein deep learning system proposed for problem. Given sequence past and durations, the estimates probabilities future their Two distinct Long-Short Term Memory (LSTM) networks are developed that cater to different assumptions about...

Identifiability is a key concern in ill-posed blind deconvolution problems arising wireless communications and image processing. The single channel version of the problem most challenging there have been efforts to use sparse models for regularizing problem. analyzed it established that simple sparsity assumption canonical basis insufficient unique recovery; surprising negative result. proof technique involves lifting into rank one matrix recovery analyzing two null-space resultant linear...

A number of ill-posed inverse problems in signal processing, like blind deconvolution, matrix factorization, dictionary learning and source separation share the common characteristic being bilinear (BIPs), i.e. observation model is a function two variables conditioned on one variable known, linear other variable. key issue that arises for such identifiability, whether sufficient to unambiguously determine pair inputs generated observation. Identifiability concern applications equalization...

We examine the problem of collecting data from an underwater sensor network using autonomous vehicle (AUV). The sensors in are equipped with acoustic modems that provide noisy, range-limited communication to AUV. One challenge this scenario is plan paths maximize information collected and minimize travel time. While executing a path, AUV can improve performance by communicating multiple nodes at once. Such multi-node requires scheduling protocol robust channel variations interference. To...

Actuated sensor networks enabled by underwater acoustic communications can be efficiently used to sense over large marine expanses that are typically challenged a paucity of resources (energy, communication bandwidth, number nodes). Many phenomena interest admit sparse representations, which, coupled with actuation and cooperation, compensate for being data starved. Herein, new methods field reconstruction, target tracking, exploration-exploitation provided, which adopt approximation,...

This paper considers identifiability and recoverability in bilinear inverse problems which is relevant to blind deconvolution matrix factorization. It shown that can be posed as rank-1 recovery subject linear constraints. Sufficient conditions for are developed the cases when rank-2 matrices present null space of operator. Signal using nuclear norm heuristic considered simple success provided.

In this paper, the estimation of a narrowband time-varying channel under practical assumptions finite block length and transmission bandwidth is investigated. It shown that signal, after passing through reveals useful parametric low-rank structure can be represented as bilinear form. To estimate channel, two strategies are developed. The first method exploits via non-convex strategy based on an alternating direction optimization between delay Doppler directions. While prior Wirtinger flow...

A problem of broad interest is the detection and localization a target or object from its generated field. In this paper, strategy which exploits structure fields designed analyzed. taking advantage structure, one able to reduce sample complexity requirements while maintaining good performance. particular, an exploration-exploitation approach proposed utilizing theory low-rank matrix completion for decaying separable The assumptions on field are fairly generic applicable many decay profiles....

Poor visibility in foggy weather stems from the fact that particles atmosphere scatter and absorb light environment reflected objects. Mathematically, de-weathering a fog degraded image is an ill posed problem existing approaches are of high complexity low versatility. In this paper, novel fuzzy logic based algorithm, to de-weather fog-degraded images, proposed. Specifically, air-light estimation carried out using followed by color correction for enhanced visibility. Experimental results...

Geo-fencing is a location based service that allows sending of messages to users who enter/exit specified geographical area, known as geo-fence. Today, it has become one the popular mobile marketing strategies. However, process designing geo-fences presently manual, i.e. retailer must specify and radius area around setup geo-fences. Moreover, this does not consider user's preference towards targeted product/service thus, can compromise his/her experience app sends these communications. We...

The detection and localization of a target from samples its generated field is problem interest in broad range applications. Often, the admits structural properties that enable design lower sample strategies with good performance. This paper designs sampling strategy which exploits separability unimodality fields theoretically analyzes trade-off achieved between density, noise level convergence rate localization. In particular, adopts an exploration-exploitation approach to utilizes theory...

A strategy for active target detection suitable the use of mobile agents in a field is presented. In particular, there an interest autonomous underwater vehicles. By exploiting notions from group testing, proposed algorithm decides when to collect new samples depending on whether agent perceives sensor measurements correspond noise or pattern. Under assumptions about emanated by target, i.e. signature locally low rank field, one can efficiently sample locate using O(m log m n) n × grid where...

Blind deconvolution is an ubiquitous non-linear inverse problem in applications like wireless communications and image processing. This generally ill-posed since signal identifiability a key concern, there have been efforts to use sparse models for regularizing blind promote identifiability. Part I of this two-part paper establishes measure theoretically tight characterization the ambiguity space unidentifiability under unconstrained inputs. II analyzes canonical-sparse surprisingly strong...

The "sum of power two (SPT)" is an effective format to represent filter coefficients in a digital which reduces the complexity multiplications filtering process just few shift and add operations. canonic SPT special sparse representation that guarantees presence at least one zero between every non-zero digits. In case adaptive filters, as are updated with time continuously, conversion such forms is, however, required each index, quite impractical requires additional circuitry. Also, position...

Detecting the presence of a target within field, is long-standing problem interest in variety applications from environmental to military. A framework considered which single autonomous underwater vehicle (AUV), physically samples decaying separable field generated by target, determine location. The overarching metric three fold trade-off between sampling and computational complexity detection, cost navigation for AUV subject low detection error probability. non-triviality this stems...

The estimation of a narrowband time-varying channel under finite block length and transmission bandwidth is investigated. A novel method proposed for in the delay-Doppler domain by exploiting structural constraints on low-rank matrix recovery. algorithm uses Gauss-Seidel iterations parameterization noisy training signal measurements. Theoretical global identifiability results leakage (due to bandwidth) are stated necessity considering Doppler shift induced structure demonstrated....

A number of important inverse problems in signal processing, including blind deconvolution, dictionary learning and matrix factorization, are instances bilinear problems. This paper shows that identifiable with probability close to one for random inputs provided the rank-2 matrices null space grows as o(mn) key applications.

Rapid growth in the consumer base for mobile services against backdrop of limited spectrum demands intelligent networks optimum utilization allocated with minimal changes existing infrastructure. This paper introduces a novel three stage predictive channel allocation scheme called Intelligent Channel Allocation (ICA) based on long term call statistics, instantaneous statistics and event driven decisions supported by cognitive radio techniques an opportunistic but fair usage spectrum. The...