Segment Anything Model (SAM) has rapidly been adopted for segmenting a wide range of natural images. However, recent studies have indicated that SAM exhibits subpar performance on 3D medical image segmentation tasks. In addition to the domain gaps between and images, disparities in spatial arrangement 2D substantial computational burden imposed by powerful GPU servers, time-consuming manual prompt generation impede extension broader spectrum applications. To mitigate these challenges, we...

Recently, bilevel optimization (BLO) has taken center stage in some very exciting developments the area of signal processing (SP) and machine learning (ML). Roughly speaking, BLO is a classical problem that involves two levels hierarchy (i.e., upper lower levels), wherein obtaining solution to upper-level requires solving lower-level one. become popular largely because it powerful modeling problems SP ML, among others, involve optimizing nested objective functions. Prominent applications...

This paper proposes a new algorithm -- the \underline{S}ingle-timescale Do\underline{u}ble-momentum \underline{St}ochastic \underline{A}pprox\underline{i}matio\underline{n} (SUSTAIN) for tackling stochastic unconstrained bilevel optimization problems. We focus on problems where lower level subproblem is strongly-convex and upper objective function smooth. Unlike prior works which rely \emph{two-timescale} or \emph{double loop} techniques, we design momentum-assisted gradient estimator both...

In this work, we consider a distributed online convex optimization problem, with time-varying (potentially adversarial) constraints. A set of nodes, jointly aim to minimize global objective function, which is the sum local functions. The and constraint functions are revealed locally at each time, after taking an action. Naturally, constraints cannot be instantaneously satisfied. Therefore, reformulate problem satisfy these in long term. To end, propose primal-dual mirror descent-based...

Federated Learning (FL) refers to the paradigm where multiple worker nodes (WNs) build a joint model by using local data. Despite extensive research, for generic non-convex FL problem, it is not clear, how choose WNs' and server's update directions, minibatch sizes, frequency, so that WNs use minimum number of samples communication rounds achieve desired solution. This work addresses above question considers class stochastic algorithms perform few updates before communication. We show when...

Adversarial training (AT) is a widely recognized defense mechanism to gain the robustness of deep neural networks against adversarial attacks. It built on min-max optimization (MMO), where minimizer (i.e., defender) seeks robust model minimize worst-case loss in presence examples crafted by maximizer attacker). However, conventional MMO method makes AT hard scale. Thus, Fast-AT (Wong et al., 2020) and other recent algorithms attempt simplify replacing its maximization step with single...

In this paper, we consider the problem of distributed sequential detection using wireless sensor networks in presence imperfect communication channels between sensors and fusion center. Sensor observations are assumed to be spatially dependent. Moreover, channel noise can dependent non-Gaussian. We propose a copula-based scheme that takes spatial dependence into account. More specifically, each local runs memory-less truncated test repeatedly sends its binary decisions center synchronously....

In this work, we propose a joint collaboration-compression framework for sequential estimation of random vector parameter in resource constrained wireless sensor network (WSN). Specifically, where the local sensors first collaborate (via collaboration matrix) with each other. Then subset selected to communicate FC linearly compress their observations before transmission. We design near-optimal and linear compression strategies under power constraints via alternating minimization minimum mean...

In this paper, we propose a distributed algorithm for stochastic smooth, non-convex optimization. We assume worker-server architecture where $N$ nodes, each having $n$ (potentially infinite) number of samples, collaborate with the help central server to perform optimization task. The global objective is minimize average local cost functions available at individual nodes. proposed approach non-trivial extension popular parallel-restarted SGD algorithm, incorporating optimal variance-reduction...

In this work, we consider the distributed stochastic optimization problem of minimizing a non-convex function $f(x) = \mathbb{E}_{ξ\sim \mathcal{D}} f(x; ξ)$ in an adversarial setting, where individual functions $f(x; can also be potentially non-convex. We assume that at most $α$-fraction total $K$ nodes Byzantines. propose robust variance-reduced gradient (SVRG) like algorithm for problem, batch gradients are computed worker (WNs) and server node (SN). For show need $\tilde{O}\left(...

The Segment Anything Model (SAM) has shown impressive performance when applied to natural image segmentation. However, it struggles with geographical images like aerial and satellite imagery, especially segmenting mobility infrastructure including roads, sidewalks, crosswalks. This inferior stems from the narrow features of these objects, their textures blending into surroundings, interference objects trees, buildings, vehicles, pedestrians - all which can disorient model produce inaccurate...

In this work, we propose a non-parametric sequential hypothesis test based on random distortion testing (RDT). RDT addresses the problem of whether or not signal, Ξ, observed in independent and identically distributed (i.i.d) additive noise deviates by more than specified tolerance, τ, from fixed model, ξ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> . The is sense that underlying signal distributions under each are assumed to be...

In this work, we propose a joint collaboration-compression framework for sequential estimation of random vector parameter in resource constrained wireless sensor network (WSN). Specifically, where the local sensors first collaborate (via collaboration matrix) with each other. Then subset selected to communicate FC linearly compress their observations before transmission. We design near-optimal and linear compression strategies under power constraints via alternating minimization minimum mean...

In this paper, we propose a new algorithm for sequential non-parametric hypothesis testing based on Random Distortion Testing (RDT). The data-based approach is in the sense that underlying signal distributions under each are assumed to be unknown. Our previously proposed non-truncated algorithm, SeqRDT, was shown achieve desired error probabilities few assumptions model. show truncated T-SeqRDT, requires even fewer model, while guaranteeing below pre-specified levels and at same time makes...

We consider the distributed stochastic optimization problem of minimizing a nonconvex function f in an adversarial setting. All w worker nodes network are expected to send their gradient vectors fusion center (or server). However, some (at most α-fraction) may be Byzantines, which arbitrary instead. Vanilla implementation descent (SGD) cannot handle such misbehavior from nodes. propose robust variant SGD is resilient presence Byzantines. The employs novel filtering rule that identifies and...

Present-day federated learning (FL) systems deployed over edge networks consists of a large number workers with high degrees heterogeneity in data and/or computing capabilities, which call for flexible worker participation terms timing, effort, heterogeneity, etc. To satisfy the need participation, we consider new FL paradigm called "Anarchic Federated Learning" (AFL) this paper. In stark contrast to conventional models, each AFL has freedom choose i) when participate FL, and ii) local steps...

As artificial intelligence is increasingly affecting all parts of society and life, there growing recognition that human interpretability machine learning models important. It often argued accuracy or other similar generalization performance metrics must be sacrificed in order to gain interpretability. Such arguments, however, fail acknowledge the overall decision-making system composed two entities: learned model a who fuses together outputs with his her own information. such, relevant...

This paper considers the problem of high dimensional signal detection in a large distributed network whose nodes can collaborate with their one-hop neighboring (spatial collaboration). We assume that only small subset communicate Fusion Center (FC). design optimal collaboration strategies which are universal for class deterministic signals. By establishing equivalence between strategy and sparse PCA, we solve efficiently evaluate impact on performance.

In this work, we propose a new method for sequential binary hypothesis testing. The approach is non-parametric in the sense that it does not assume any knowledge of signal distributions under each hypothesis. proposed framework based on Random distortion testing (RDT) which addresses problem whether or random signal, deviates by more than specified tolerance, $\tau$, from fixed value, $\xi_{0}$. We first state setup and then discuss earlier approaches to solve problem. algorithm, T-SeqRDT,...