Despite recent advances, vision-language models trained with standard contrastive objectives still struggle with compositional reasoning -- the ability to understand structured relationships between visual and linguistic elements. This shortcoming is largely due to the tendency of the text encoder to focus on individual words rather than their relations, a limitation reinforced by contrastive training that primarily aligns words with visual objects. In this paper, we introduce REconstruction and Alignment of text Descriptions (READ), a fine-tuning method designed to enhance compositional reasoning by adding two auxiliary objectives to the contrastive learning: (1) a token-level reconstruction objective, where a frozen pre-trained decoder reconstructs paraphrased captions based on the embedding of the original caption; and (2) a sentence-level alignment objective, which explicitly aligns paraphrased sentences in the embedding space. We show that READ-CLIP, a model derived by applying the READ method to the pre-trained CLIP model, achieves the state-of-the-art performance across five major compositional reasoning benchmarks, outperforming the strongest conventional fine-tuning baseline by up to 4.1%. Furthermore, applying READ to existing CLIP variants (including NegCLIP and FSC-CLIP) also improves performance on these benchmarks. Quantitative and qualitative analyses reveal that our proposed objectives -- reconstruction and alignment -- offer complementary benefits: the former encourages the encoder to capture relationships between words within a caption, while the latter ensures consistent representations for paraphrases expressed with different wording.

Post-quantum cryptography (PQC) is becoming increasingly critical for securing Internet of Things (IoT) applications against potential threats posed by quantum computers. However, cryptographic schemes based on mathematically hard problems, such as lattice-based constructions, may face reduced security margins due to ongoing cryptanalytic advancements. In IoT use cases requiring long-term deployment without guaranteed secure updates, symmetric-based signature schemes, such as$\mathsf {SPHINCS+}$, whose security relies solely on symmetric primitives, are considered robust alternatives. Among these,$\textsf {AIMer}$, selected as a finalist in the KpqC competition, is emerging as a promising candidate. This article presents optimized software implementations of$\textsf {AIMer}$for various IoT platforms, ranging from low-end to high-end devices. We propose memory-optimized and time–memory tradeoff implementations for resource-constrained ARM Cortex-M4 devices, and single instruction multiple data (SIMD)-accelerated implementations for AArch64 and AVX2 architectures. Furthermore, we conduct a comprehensive performance evaluation comparing our optimized implementations with other PQC signature schemes. Our results show that$\textsf {AIMer}$achieves significantly smaller signature sizes and faster key generation and signing compared to$\mathsf {SPHINCS+}$across all security levels, making it a viable option for long-term IoT deployment.

Post-quantum signature schemes based on the MPC-in-the-Head (MPCitH) paradigm are recently attracting significant attention as their security solely depends on the one-wayness of the underlying primitive, providing diversity for the hardness assumption in post-quantum cryptography. Recent MPCitH-friendly ciphers have been designed using simple algebraic S-boxes operating on a large field in order to improve the performance of the resulting signature schemes. Due to their simple algebraic structures, their security against algebraic attacks should be comprehensively studied.

Along with the resistance against quantum computers, isogeny-based cryptography offers attractive cryptosystems due to small key sizes and compatibility with the current elliptic curve primitives. While the state-of-the-art implementation uses Montgomery curves, which facilitates efficient elliptic curve arithmetic and isogeny computations, other forms of elliptic curves can be used to produce an efficient result. In this paper, we present the new hybrid method for isogeny-based cryptosystem using Edwards curves. Unlike the previous hybrid methods, we exploit Edwards curves for recovering the curve coefficients and Montgomery curves for other operations. To this end, we first carefully examine and compare the computational cost of Montgomery and Edwards isogenies. Then, we fine-tune and tailor Edwards isogenies in order to blend with Montgomery isogenies efficiently. Additionally, we present the implementation results of Supersingular Isogeny Diffie-Hellman (SIDH) key exchange using the proposed method. We demonstrate that our method outperforms the previously proposed hybrid method, and is as fast as Montgomery-only implementation. Our results show that proper use of Edwards curves for isogeny-based cryptosystem can be quite practical.

The isogeny-based cryptosystem is the most recent category in the field of postquantum cryptography. However, it is widely studied due to short key sizes and compatibility with the current elliptic curve primitives. The main building blocks when implementing the isogeny-based cryptosystem are isogeny computations and point operations. From isogeny construction perspective, since the cryptosystem moves along the isogeny graph, isogeny formula cannot be optimized for specific coefficients of elliptic curves. Therefore, Montgomery curves are used in the literature, due to the efficient point operation on an arbitrary elliptic curve. In this paper, we propose formulas for computing 3 and 4 isogenies on twisted Edwards curves. Additionally, we further optimize our isogeny formulas on Edwards curves and compare the computational cost of Montgomery curves. We also present the implementation results of our isogeny computations and demonstrate that isogenies on Edwards curves are as efficient as those on Montgomery curves.

In this letter, we propose a method for detecting a human subject using Doppler radar by investigating the physical characteristics of targets. Human detection has a number of applications in security, surveillance, and search-and-rescue operations. To classify a target from the Doppler signal, several features related to the physical characteristics of a target are extracted from a spectrogram. The features include the frequency of the limb motion, stride, bandwidth of the Doppler signal, and distribution of the signal strength in a spectrogram. The main contribution of this letter is the use of stride information of a target for the classification. Owing to the different lengths of legs and kinematic signatures of the target species, a human subject occupies a unique space in the domain of the stride and the frequency of limb motion. To verify the proposed method, we investigated humans, dogs, bicycles, and vehicles using the developed continuous-wave Doppler radar. The human subject is identified by a classifier of a support vector machine (SVM) trained to the extracted features. The trained SVM can detect a human subject with an accuracy of 96% with fourfold cross validation.