Sentiment Analysis (SA) classifies text into positive, negative, or neutral categories and is widely used in social media moderation, market research, and behavioral analytics. However, most existing models struggle with multilingual and code-mixed content, especially Tamil-English text that includes transliteration inconsistencies, informal spellings, and limited linguistic resources. To address this gap, this research introduces a Multi-granular Token-level Heterogeneous Graph transformer with Adaptive Revolution Optimization (MTHG-ARO). The system begins with language-aware preprocessing, including token-level language identification, transliteration correction, normalization, and code-mix ratio estimation to reduce noise. Next, multilingual embeddings are generated using IndicBERT (for Tamil) and FastText (for English/German), ensuring cross-lingual semantic alignment. Deeper contextual understanding is achieved through the Cross-Lingual Feature-Attentive Linguistic Signal Encoder (CL-FALSE), which models inter-language relationships, while the Adaptive Multi-modal Gated Fusion with Prompt Learning (AMGF-PL) selectively emphasizes important linguistic cues. Training stability is maintained by the Adaptive Revolution Optimizer (ARO), which uses dynamic inertia control and controlled perturbation to avoid gradient stagnation and improve convergence. Experiments on the DravidianCodeMix-Tamil (DCM-T), German Hate Speech (GHS), and YouTube Comments (YTC) datasets show that MTHG-ARO achieves an average accuracy of 97.83
Burst super-resolution (BSR) achieves higher reconstruction fidelity than single-image super-resolution by aggregating sub-pixel information from multiple frames. However, it requires a burst of frames captured under near-static conditions, and such frames are unavailable for single photographs or archival images. We propose Pseudo-Burst Super-Resolution (PBSR), which brings BSR-style reconstruction to the single-image input setting by synthesizing pseudo-burst frames via zero-shot novel view synthesis (NVS). PBSR first generates pseudo-burst frames with realistic parallax using a viewpoint sampling strategy calibrated to the pixel-shift distribution of real burst datasets. The frames are then processed by a BSR backbone equipped with a geometry-aware alignment module that reuses the NVS camera parameters and estimated depth. While the NVS stage is zero-shot, the backbone is fine-tuned with an artifact-aware adaptation so that it can exploit the synthesized frames despite their generative artifacts. Experiments on the BurstSR dataset show that PBSR consistently improves in PSNR, SSIM, and LPIPS over the single-frame baseline with the same reconstruction backbone.
Card-based protocols perform secure computation using physical playing cards, and their efficiency is primarily evaluated by two metrics: the number of cards and the number of shuffles. In this paper, we present three new card-based protocols that reduce the number of cards. The first protocol supports any Boolean circuit and decreases the number of cards. The second protocol extends the first to a committed-format version, while the third is specialized for Boolean circuits without a branched input, achieving an even smaller number of cards. All the proposed protocols are based on Yao’s garbled-circuit framework, similar to previous single-shuffle card-based protocols. Our main contribution is a novel encoding method that represents the truth table of each Boolean gate with fewer cards than existing schemes. Consequently, our protocols require only six cards per gate in a single-shuffle execution, providing a significant improvement in efficiency over prior work. Furthermore, by incorporating the card-based free-XOR introduced by Manabe–Shinagawa, we can further reduce the number of cards in certain configurations.
In this study, we explore visual prompt injection (VPI), a method that exploits the ability of large vision-language models (LVLMs) to follow instructions embedded within input images. LVLMs, such as GPT-4V, are designed to interpret and respond to visual prompts. Although this capability enables valuable applications, it introduces substantial security risks. To address this issue, we propose a new VPI method called “goal hijacking via visual prompt injection” (GHVPI). This method redirects LVLMs from their original execution task to an alternative task specified by an attacker. Our quantitative analysis reveals that GPT-4V is vulnerable to GHVPI, with an attack success rate of 15.8
This paper provides an empirical study of feature learning based on induction, presented as a foundational proof-of-concept for bridging raw sensory input and symbolic reasoning. We first encode image data into first-order expression. We then compute their least generalization to extract a common pattern of input data. Next, we introduce two different methods for feature learning. The frequency method extracts pixel values that appear most frequently in each location. The average method computes the average values of the pixels in each location. They are also applied to convolutional methods for data with random location. We perform experiments using the MNIST and CIFAR-10 datasets and show that the proposed methods successfully capture features from training data and classify test data in about 90
