Binary Neural Networks (BNNs) are receiving an up-surge of attention for bringing power-hungry deep learning towards edge devices. The traditional wisdom in this space is to employ sign(.) for binarizing feature maps. We argue and illustrate that sign(.) is a uniqueness bottleneck, limiting information propagation throughout the network. To alleviate this, we propose to dispense sign(.), replacing it with a learnable activation binarizer (LAB), allowing the network to learn a fine-grained binarization kernel per layer - as opposed to global thresholding. LAB is a novel universal module that can seamlessly be integrated into existing architectures. To confirm this, we plug it into four seminal BNNs and show a considerable accuracy boost at the cost of tolerable increase in delay and complexity. Finally, we build an end-to-end BNN (coined as LAB-BNN) around LAB, and demonstrate that it achieves competitive performance on par with the state-of-the-art on ImageNet. Our code can be found in our repository: https://github.com/sfalkena/LAB.@en

Humans have a unique ability to learn new representations from just a handful of examples with little to no supervision. Deep learning models, however, require an abundance of data and supervision to perform at a satisfactory level. Unsupervised few-shot learning (U-FSL) is the pursuit of bridging this gap between machines and humans. Inspired by the capacity of graph neural networks (GNNs) in discovering complex inter-sample relationships, we propose a novel self-attention based message passing contrastive learning approach (coined as SAMP-CLR) for U-FSL pre-training. We also propose an optimal transport (OT) based fine-tuning strategy (we call OpT-Tune) to efficiently induce task awareness into our novel end-to-end unsupervised few-shot classification framework (SAMPTransfer). Our extensive experimental results corroborate the efficacy of SAMPTransferin a variety of downstream few-shot classification scenarios, setting a new state-of-the-art for U-FSL on both miniImageNet and tieredImageNet benchmarks, offering up to 7%+ and 5%+ improvements, respectively. Our further investigations also confirm that SAMPTransferremains on-par with some supervised baselines on miniImageNet and outperforms all existing U-FSL baselines in a challenging cross-domain scenario. Our code can be found in our GitHub repository: https://github.com/ojss/SAMPTransfer/.@en

Unsupervised learning is argued to be the dark matter of human intelligence. To build in this direction, this paper focuses on unsupervised learning from an abundance of unlabeled data followed by few-shot fine-tuning on a downstream classification task. To this aim, we extend a recent study on adopting contrastive learning for self-supervised pre-training by incorporating class-level cognizance through iterative clustering and re-ranking and by expanding the contrastive optimization loss to account for it. To our knowledge, our experimentation both in standard and cross-domain scenarios demonstrate that we set a new state-of-the-art (SoTA) in (5-way, 1 and 5-shot) settings of standard mini-ImageNet benchmark as well as the (5-way, 5 and 20-shot) settings of cross-domain CDFSL benchmark. Our code and experimentation can be found in our GitHub repository: https://github.com/ojss/c3lr.

@en

Classical federated learning approaches incur significant performance degradation in the presence of non-independent and identically distributed (non-IID) client data. A possible direction to address this issue is forming clusters of clients with roughly IID data. Most solutions following this direction are iterative and relatively slow, also prone to convergence issues in discovering underlying cluster formations. We introduce federated learning with taskonomy (FLT) that generalizes this direction by learning the task relatedness between clients for more efficient federated aggregation of heterogeneous data. In a one-off process, the server provides the clients with a pretrained (and fine-tunable) encoder to compress their data into a latent representation and transmit the signature of their data back to the server. The server then learns the task relatedness among clients via manifold learning and performs a generalization of federated averaging. FLT can flexibly handle a generic client relatedness graph, when there are no explicit clusters of clients, as well as efficiently decompose it into (disjoint) clusters for clustered federated learning. We demonstrate that FLT not only outperforms the existing state-of-the-art baselines in non-IID scenarios but also offers improved fairness across clients. Our codebase can be found at: https://github.com/hjraad/FLT/

@en

Traditional empirical risk minimization (ERM) for semantic segmentation can disproportionately advantage or disadvantage certain target classes in favor of an (unfair but) improved overall performance. Inspired by the recently introduced tilted ERM (TERM), we propose tilted cross-entropy (TCE) loss and adapt it to the semantic segmentation set-ting to minimize performance disparity among target classes and promote fairness. Through quantitative and qualitative performance analyses, we demonstrate that the proposed Stochastic TCE for semantic segmentation can offer improved overall fairness by efficiently minimizing the performance disparity among the target classes of Cityscapes.

@en

Semantic segmentation is one of the most fundamental problems in computer vision with significant impact on a wide variety of applications. Adversarial learning is shown to be an effective approach for improving semantic segmentation quality by enforcing higher-level pixel correlations and structural information. However, state-of-the-art semantic segmentation models cannot be easily plugged into an adversarial setting because they are not designed to accommodate convergence and stability issues in adversarial networks. We bridge this gap by building a conditional adversarial network with a state-of-the-art segmentation model (DeepLabv3+) at its core. To battle the stability issues, we introduce a novel lookahead adversarial learning (LoAd) approach with an embedded label map aggregation module. We focus on semantic segmentation models that run fast at inference for near real-time field applications. Through extensive experimentation, we demonstrate that the proposed solution can alleviate divergence issues in an adversarial semantic segmentation setting and results in considerable performance improvements (+5% in some classes) on the baseline for three standard datasets.

@en

This paper focuses on the design of a Fourier dictionary matrix formed by selecting specific rows of the inverse discrete Fourier transform matrix based on coherence-related metrics. While maximum coherence is a popular metric in compressive sampling, we also consider rms LN-coherence, which focuses on the largest LN (instead of one) inner products between different columns of the dictionary matrix. Finding a dictionary matrix optimizing either the maximum or the rms LN-coherence lead to a complicated optimization problem. Hence, we introduce a new metric called coherence deviation (CD), which gives a measure on the variation of all the inner products between different columns of the dictionary matrix, and motivate its use as an amenable alternative for both the maximum and rms LN-coherence. While finding a dictionary matrix optimizing the CD leads to a simplified optimization problem, the resulting cost function is a quartic function of a binary vector variable. Hence, we propose Greedy-β algorithm to provide sub-optimal solutions.@en

Dual decomposition has been successfully employed in a variety of distributed convex optimization problems solved by a network of computing and communicating nodes. Often, when the cost function is separable but the constraints are coupled, the dual decomposition scheme involves local parallel subgradient calculations and a global subgradient update performed by a master node. In this paper, we propose a consensus-based dual decomposition to remove the need for such a master node and still enable the computing nodes to generate an approximate dual solution for the underlying convex optimization problem. In addition, we provide a primal recovery mechanism to allow the nodes to have access to approximate near-optimal primal solutions. Our scheme is based on a constant stepsize choice, and the dual and primal objective convergence are achieved up to a bounded error floor dependent on the stepsize and on the number of consensus steps among the nodes.@en

INSPEC Accession Number: 13820978@en