In many industry scale applications, large and resource consuming machine learning models reside in powerful cloud servers. At the same time, amounts of input data are collected at edge cloud. The inference results also communicated to users or passed downstream tasks edge. often consists a number low-power devices. It is big challenge design products support sophisticated deep model deployment conduct an efficient manner so that accuracy remains high end-to-end latency kept low. This paper...

The main challenge in domain generalization (DG) is to handle the distribution shift problem that lies between training and test data. Recent studies suggest test-time (TTT), which adapts learned model with data, might be a promising solution problem. Generally, TTT strategy hinges its performance on two factors: selecting an appropriate auxiliary task for updating identifying reliable parameters update during phase. Both previous arts our experiments indicate may not improve but detrimental...

Many vision-based applications rely on logistic regression for embedding classification within a probabilistic context, such as recognition in images and videos or identifying disease-specific image phenotypes from neuroimages. Logistic regression, however, often performs poorly when trained data that is noisy, has irrelevant features, the samples are distributed across classes an imbalanced setting; common occurrence visual tasks. To deal with those issues, researchers generally ad-hoc...

Diffusion magnetic resonance imaging (DMRI) suffers from lower signal-to-noise-ratio (SNR) due to MR signal attenuation associated with the motion of water molecules. To improve SNR, non-local means (NLM) algorithm has demonstrated state-of-the-art performance in noise reduction. However, existing NLM algorithms do not take into account explicitly fact that DMRI can vary significantly local fiber orientations. Applying naïvely hence blur subtle structures and aggravate partial volume...

HIV-Associated Neurocognitive Disorder (HAND) is the most common constellation of cognitive dysfunctions in chronic HIV infected patients age 60 or older U.S. Only few published methods assist distinguishing HAND from other forms age-associated decline, such as Mild Cognitive Impairment (MCI). In this report, a data-driven, nonparameteric model to identify morphometric patterns separating MCI due non-HIV conditions group was proposed. This enhanced potential for separation by combining...

Many model watermarking methods have been developed to prevent valuable deployed commercial models from being stealthily stolen by distillations. However, watermarks produced most existing can be easily evaded ensemble distillation, because averaging the outputs of multiple ensembled significantly reduce or even erase watermarks. In this paper, we focus on tackling challenging task defending against distillation. We propose a novel technique named CosWM achieve outstanding performance is not...

Contrastive self-supervised representation learning methods maximize the similarity between positive pairs, and at same time tend to minimize negative pairs. However, in general interplay pairs is ignored as they do not put place special mechanisms treat differently according their specific differences similarities. In this paper, we present Extended Momentum Contrast (XMoCo), a method founded upon legacy of momentum-encoder unit proposed MoCo family configurations. To end, introduce cross...

In today's digital world, enterprises and individuals are generating massive data that is potentially useful for many consumers with driven applications. The emergence of marketplaces a step toward helping the owners to monetize their assets get connected potential buyers. current cannot handle challenges related ownership claims, illegal redistribution, traceability. To overcome these problems in general-purpose market, we propose marketplace based on watermarking Non-Fungible Token (NFT)...

Branch-and-bound is a systematic enumerative method for combinatorial optimization, where the performance highly relies on variable selection strategy. State-of-the-art handcrafted heuristic strategies suffer from relatively slow inference time each selection, while current machine learning methods require significant amount of labeled data. We propose new approach solving data labeling and latency issues in optimization based use reinforcement (RL) paradigm. imitation to bootstrap an RL...

Diffusion MRI requires sufficient coverage of the diffusion wavevector space, also known as q-space, to adequately capture pattern water in various directions and scales. As a result, acquisition time can be prohibitive for individuals who are unable stay still scanner an extensive period time, such infants. To address this problem, paper we harness non-local self-similar information x-q space data q-space upsampling. Specifically, first perform neighborhood matching establish relationships...

The Natural Language for Optimization (NL4Opt) Competition was created to investigate methods of extracting the meaning and formulation an optimization problem based on its text description. Specifically, goal competition is increase accessibility usability solvers by allowing non-experts interface with them using natural language. We separate this challenging into two sub-tasks: (1) recognize label semantic entities that correspond components problem; (2) generate a representation (i.e.,...

Building huge and highly capable language models has been a trend in the past years. Despite their great performance, they incur high computational cost. A common solution is to apply model compression or choose light-weight architectures, which often need separate fixed-size for each desirable budget, may lose performance case of heavy compression. This paper proposes an effective dynamic inference approach, called E-LANG, distributes between large accurate Super-models Swift models. To...

Semi-supervised learning (SSL) has seen great strides when labeled data is scarce but unlabeled abundant. Critically, most recent work assume that such drawn from the same distribution as data. In this work, we show state-of-the-art SSL algorithms suffer a degradation in performance presence of auxiliary does not necessarily possess class set. We term problem Auxiliary-SSL and propose AuxMix, an algorithm leverages self-supervised tasks to learn generic features order mask are semantically...

Decentralized optimization, particularly the class of decentralized composite convex optimization (DCCO) problems, has found many applications. Due to ubiquitous communication congestion and random dropouts in practice, it is highly desirable design algorithms that can handle stochastic networks. However, most existing for DCCO only work networks are deterministically connected during bounded rounds, therefore, cannot be extended In this article, we propose a new dual averaging (DDA)...

Diffusion MRI derives its contrast from MR signal attenuation induced by the movement of water molecules in microstructural environments. Associated with is reduction signal-to-noise ratio (SNR). Methods based on total variation (TV) have shown superior performance image noise reduction. However, TV denoising can result stair-casing effects due to inherent piecewise-constant assumption. In this paper, we propose a tight wavelet frame approach for edge-preserving diffusion-weighted (DW)...

Curating high quality datasets that play a key role in the emergence of new AI applications requires considerable time, money, and computational resources. So, effective ownership protection is becoming critical. Recently, to protect an image dataset, imperceptible watermarking techniques are used store information (i.e., watermark) into individual samples. Embedding entire watermark all samples leads significant redundancy embedded which damages watermarked dataset extraction accuracy. In...

The main challenge in domain generalization (DG) is to handle the distribution shift problem that lies between training and test data. Recent studies suggest test-time (TTT), which adapts learned model with data, might be a promising solution problem. Generally, TTT strategy hinges its performance on two factors: selecting an appropriate auxiliary task for updating identifying reliable parameters update during phase. Both previous arts our experiments indicate may not improve but detrimental...