Search

Federated Learning (FL) for face recognition aggregates locally optimized models from individual clients to construct a generalized face recognition model. However, previous studies present two major challenges: insufficient incorporation of self-supervised learning and the necessity for clients to accommodate multiple subjects. To tackle these limitations, we propose FedFS (Federated Learning for personalized Face recognition via intra-subject Self-supervised learning framework), a novel federated learning architecture tailored to train personalized face recognition models without imposing subjects. Our proposed FedFS comprises two crucial components that leverage aggregated features of the local and global models to cooperate with representations of an off-the-shelf model. These components are (1) adaptive soft label construction, utilizing dot product operations to reformat labels within intra-instances, and (2) intra-subject self-supervised learning, employing cosine similarity operations to strengthen robust intra-subject representations. Additionally, we introduce a regularization loss to prevent overfitting and ensure the stability of the optimized model. To assess the effectiveness of FedFS, we conduct comprehensive experiments on the DigiFace-1M and VGGFace datasets, demonstrating superior performance compared to previous methods., Comment: Accepted at the The 35th British Machine Vision Conference 2024 (BMVC 2024), Glasgow, UK. Youngjun Kwak is corresponding author

Large-scale quantum circuits are required to exploit the advantages of quantum computers. Present-day quantum computers have become less reliable with increasing depths of quantum circuits. To overcome this limitation, quantum error-correction codes have been introduced. Although the success of quantum error correction codes has been announced in Google[1, 2] and neutral atom[3] quantum computers, there have been no reports on IBM quantum computers showing error suppression owing to its unique heavy-hexagon structure. This structure restricts connectivity, and quantum error-correction codes on IBM quantum computers require flag qubits. Here, we report the successful implementation of a syndrome extraction circuit with flag qubits on IBM quantum computers. Moreover, we demonstrate its effectiveness by considering the repetition code as a test code among the quantum error-correcting codes. Even though the data qubit is not adjacent to the syndrome qubit, logical error rates diminish exponentially as the distance of the repetition code increases from three to nine. Even when two flag qubits exist between the data and syndrome qubits, the logical error rates decrease as the distance increases similarly. This confirms the successful implementation of the syndrome extraction circuit with flag qubits on the IBM quantum computer.

For a quadratic field $\mathcal{K}$ without rationally defined CM, we prove that there exists of a prime $p_{\mathcal{K}}$ depending only on $\mathcal{K}$ such that if $d$ is a positive integer whose minimal prime divisor is greater than $p_{\mathcal{K}}$, then for any extension $L/\mathcal{K}$ of degree d and any elliptic curve $E/\mathcal{K}$, we have $E\left(L\right)_{\operatorname{tors}} = E\left(\mathcal{K}\right)_{\operatorname{tors}}$. By not assuming the GRH, this is a generalization of the results by Genao, and Gon\'alez-Jim\'enez and Najman., Comment: The title and typos are corrected

Federated learning (FL) is a promising approach for enhancing data privacy preservation, particularly for authentication systems. However, limited round communications, scarce representation, and scalability pose significant challenges to its deployment, hindering its full potential. In this paper, we propose 'ProtoFL', Prototypical Representation Distillation based unsupervised Federated Learning to enhance the representation power of a global model and reduce round communication costs. Additionally, we introduce a local one-class classifier based on normalizing flows to improve performance with limited data. Our study represents the first investigation of using FL to improve one-class classification performance. We conduct extensive experiments on five widely used benchmarks, namely MNIST, CIFAR-10, CIFAR-100, ImageNet-30, and Keystroke-Dynamics, to demonstrate the superior performance of our proposed framework over previous methods in the literature., Comment: Accepted by ICCV 2023. Hansol Kim and Youngjun Kwak contributed equally to this work

A distributed control of vehicle platooning is referred to as distributed consensus (DC) since many autonomous vehicles (AVs) reach a consensus to move as one body with the same velocity and inter-distance. For DC control to be stable, other AVs' real-time position information should be inputted to each AV's controller via vehicle-to-vehicle (V2V) communications. On the other hand, too many V2V links should be simultaneously established and frequently retrained, causing frequent packet loss and longer communication latency. We propose a novel DC algorithm called over-the-air consensus (AirCons), a joint communication-and-control design with two key features to overcome the above limitations. First, exploiting a wireless signal's superposition and broadcasting properties renders all AVs' signals to converge to a specific value proportional to participating AVs' average position without individual V2V channel information. Second, the estimated average position is used to control each AV's dynamics instead of each AV's individual position. Through analytic and numerical studies, the effectiveness of the proposed AirCons designed on the state-of-the-art New Radio architecture is verified by showing a $14.22\%$ control gain compared to the benchmark without the average position., Comment: This work has been submitted to the IEEE for possible publication

Let $K$ be a number field. For positive integers $m$ and $n$ such that $m\mid n$, we let $\mathscr{S}_{m,n}$ be the set of elliptic curves $E/K$ defined over $K$ such that $E(K)_{\operatorname{tors}}\supseteq \mathscr{T}\cong \mathbb{Z}/m\mathbb{Z}\times \mathbb{Z}/n\mathbb{Z}$. We prove that if the genus of the modular curve $X_{1}(m,n)$ is $0$, then `almost all' $E\in \mathscr{S}_{m,n}$ satisfy that $E(K)_{\operatorname{tors}}= \mathscr{T}$, i.e., no larger than $\mathscr{T}$. In particular, if $m=n=1$, this result generalizes Duke's theorem over $\mathbb{Q}$ to arbitrary number fields $K$ for the trivial torsion subgroup., Comment: Corollary 1.8 is added and more explanation about the raised questions are added in the introduciton

We give a classification of maximal elements of the set of finite groups that can be realized as the automorphism groups of polarized abelian threefolds over finite fields., Comment: Any comments will be appreciated. arXiv admin note: substantial text overlap with arXiv:1809.06251, arXiv:2002.05535