Abstract

We present CURIO, an OCR system for low-resource
historical manuscripts. In many challenging cases,
manuscripts feature curved text lines, unsegmented lines
with lack of spacing between words, and highly vari-
able line lengths -- conditions under which existing
OCR methods fail. To tackle this challenge, we first
extract lines and corresponding curvature profiles from
manuscripts, then straighten them using a rectification
procedure to reduce redundant background within each
line. Because data is scarce, we compliment real data
with synthetic data. To bridge the synthetic-real gap,
we generate line images by warping rendered straight
text along the rectified profiles, ensuring both real and
synthetic lines align in their curvature characteristics.
Our recognizer is a lightweight CNN-Transformer with
padding-aware null activations, sparse attention and
optimized with CTC loss for efficient training. We
evaluate our method on challenging manuscript collec-
tions written in Sharada, a rare and endangered Indic
script. CURIO outperforms strong CNN+RNN and
Transformer baselines, with the largest gains on high-
curvature lines and long lines. CURIO further transfers
zero-shot to printed Sharada text, indicating robustness
beyond manuscript domain.

Abstract

Foundation models for time series forecasting are highly sensitive to configuration parameters such as context length and patch size, which substantially influence predictive performance. These parameters are typically chosen via static defaults or per-series hyperparameter search using AutoML, the latter requiring repeated evaluations of a large model. We reinterpret configuration selection as an amortized learning problem: instead of optimizing configurations independently for each new series, we learn a lightweight learning-to-rank model that predicts high-performing configurations. Using the Moirai 1.1-R-small foundation model on the Uber TLC and Electricity benchmarks, Tune-as-Inference achieves accuracy within 3-5% of 20-trial Bayesian optimization while reducing per-series configuration time by approximately 20 times. These results suggest that configuration adaptation for time series foundation models are better treated as inference rather than iterative search.

Abstract

Product reviews on e-commerce platforms are a critical form of user-generated content that influence consumer decisions. However, these reviews are predominantly in English, creating a significant accessibility barrier for users who are not fluent in English. When translating into major Indian languages using the current models, the outputs often fail to capture domain-specific features and colloquial style, resulting in stylistically unnatural texts. To address this gap, we introduce textbf{STAR-IL}, a human-annotated, multilingual, parallel corpus for style-aware translation of product reviews. We evaluate the performance of several state-of-the-art models on our dataset for the task of product review translation. Our experiments show that models fine-tuned on STAR-IL achieve significant average performance gain of textbf{5.77} points in BLEU and textbf{3.78} points in COMET, when compared to their baselines, across all languages. Our dataset provides a valuable benchmark for future research in style-aware product review translation. The STAR-IL dataset is publicly available at https://github.com/ltrc/STAR-IL-Corpus

Abstract

The extent to which large language models (LLMs) can perform culturally grounded reasoning across non-English languages remains underexplored. This paper examines the reasoning and self-assessment abilities of LLMs across seven major Indian languages- Bengali, Gujarati, Hindi, Kannada, Malayalam, Tamil, and Telugu. We introduce a multilingual riddle dataset combining traditional riddles with context-reconstructed variants and evaluate five LLMs- Gemini 2.5 Pro, Gemini 2.5 Flash, Mistral-Saba, LLaMA-4-Scout, and LLaMA-4-Maverick under seven prompting strategies. In the first stage, we assess riddle-solving performance and find that while Gemini 2.5 Pro performs best overall, few-shot methods yield only marginal gains, and accuracy varies notably across languages. In the second stage, we conduct a self-evaluation experiment to measure reasoning consistency. The results reveal a key finding: a model's initial accuracy is inversely correlated with its ability to identify its own mistakes. Top-performing models such as Gemini 2.5 Pro are overconfident (4.34% True Negative Rate), whereas lower-performing models like LLaMA-4-Scout are substantially more self-aware (42.09% True Negative Rate). These results point to clear gaps in multilingual reasoning and highlight the need for models that not only reason effectively but also recognize their own limitations.

Abstract

This study analyses the slope stability of hills under different heights of retaining walls and slope angles, both with and without water pressure, using the Morgenstern-Price analysis method in GeoStudio software. The slip surface option employed is Entry and Exit, and the slope stability material model is Mohr-Coulomb. The deciding parameter for the analysis is the Factor of Safety. The study finds that when water pressure is considered, the stability of slopes with the same angle and wall height is lower than that of slopes without water pressure. Additionally, as the slope angle increases, the stability of the slope decreases. The study also reveals that a slope that is stable at 30˚ under no water pressure condition becomes unstable when water pressure is considered. Finally, based on the available data and analysis results, the study infers that more than half of the area is considered unsafe since slopes with a gradient of more than 30˚ have a factor of safety of less than one. The findings of this study can be used to design and plan effective slope stabilization measures and land use planning in the region.

Abstract

Digital fitness technologies promise greater awareness, structured routines, and enhanced discipline, influencing how individuals construct notions of well-being and engage in fitness within social environments. In urban India, these dynamics are shaped by infrastructural constraints, uneven access to fitness resources, and context-specific conditions that complicate standardized tracking models built by global platforms. Drawing on prior literature studies and qualitative fieldwork, we examine how young fitness practitioners adopt and negotiate the use of self-tracking technologies. Using bricolage as an analytical lens, we suggest how users creatively repurpose tools, navigating data-driven guidance and embodied intuition. Our findings show three key themes: creation of hybrid self-tracking ecologies, technology domestication via selective appropriation, and adaptation of fitness tracking to local sociomaterial contexts. Fitness emerges as a situated and relational practice shaped by resource constraints and interpretive self-tracking.

Abstract

This paper presents a 13.95-16.96 GHz low power, low phase noise (PN) LC voltage-controlled oscillator (VCO) in 65 nm CMOS technology.
To address the limitations of conventional harmonic tuned oscillators, whose phase noise (PN) degrades significantly across the tuning range (TR) due to narrowband harmonic impedances, a custom transformer is designed and analyzed.
The proposed tank can actively contribute to ISF shaping and achieve consistent flicker noise upconversion reduction across the entire tuning range. Post-layout simulations demonstrate a phase noise (PN) of -118 dBc/Hz and a figure of merit (FoM) exceeding 187 dBc/Hz at a 1 MHz offset. The VCO occupies only 0.0306 mm$^2$, and its PN performance makes it suitable for mm-Wave applications and next-generation wireless systems.

Abstract

Large Language Models (LLMs) excel on English reasoning tasks but falter on morphologically rich, low-resource languages such as Telugu, Tamil, and Kannada. We present TeluguEval, a human-curated reasoning benchmark created by translating GSM8K (math), Winogrande (commonsense), ARC (science), CaseHOLD (law), and Hendrycks Ethics into Telugu. We evaluate eight models spanning global (Llama-3.1-8B, Llama-2-7B, Qwen-8B, Gemma-7B, Gemini-2.0) and regional (Telugu-Llama2-7B, Indic-Gemma-7B, Sarvam-m-24B) systems. While extremely strong models such as Gemini and Sarvam-m largely retain performance in Telugu, most English-centric models suffer severe accuracy drops, often exceeding 30 to 40 points, particularly on mathematical and scientific reasoning. We further observe systematic failure modes including script sensitivity, option-selection bias, repetition loops, and unintended code-switching. Our results demonstrate that surface-level Telugu fluency does not imply robust reasoning capability, underscoring the need for Telugu-specific data, tokenization, and pretraining. TeluguEval provides a standardized testbed to drive progress on reasoning in low-resource Indian languages.

Abstract

Degradation of ECG electrodes critically impacts the reliability of long-term cardiac monitoring. Existing systems typically fail to detect electrode deterioration until signal quality has already degraded. This paper presents an impedance-based electrode quality assessment system that operates along with ECG system. The system estimates the electrode time constant (τ) using RC decay analysis and jointly evaluates τ and interface capacitance (Cint) to identify electrode degradation. Commercial Ag/AgCl electrodes were evaluated under simulated thermal, saline, and mechanical stress conditions. Baseline parameters for the device under test (DUT), obtained by averaging measure- ments over 56,000 excitation cycles per minute, were τ0 = 0.103± 0.018 ms and Cint,0 = 57.27 nF, while degraded state values reached τmax = 1.151 ± 0.112 ms and Cint,max = 2071.33 nF. A degradation threshold of τ > 0.206 ms (corresponding to > 5σ above the baseline) was established to generate a warning to the user. Implemented on an ESP32-C3, the system consumes 7.7 mW and provides a low-cost, real-time method for monitoring electrode health. The proposed scheme is readily adaptable to a wide range of physiological sensing systems.

Abstract

High-dimensional datasets present a fundamental challenge:
understanding how data points relate to each other across different feature combinations. Existing visualization methods typically show global structure through dimensionality reduction or highlight relationships in the full feature space, but they fail to reveal which specific feature subspaces drive point-to-point connections. We present CSP-HEIDI, a visualization framework that resolves this ambiguity by analyzing knearest neighbor (k-NN) relationships across the complete feature subspaces. Central to our approach is the Interpretable Expression Matrix (IEM), which replaces scalar distances with minimal Boolean expressions (minterms). By employing a logic minimization strategy, CSPHEIDI isolates the "root cause" subspaces that are sufficient to maintain neighborhood connections, discarding redundant supersets. Through a novel Concentric Spiral Arrangement, these expressions are organized to reveal structural topology. We demonstrate on biological and synthetic datasets that CSP-HEIDI successfully disentangles intra-cluster cohesion from inter-cluster bridging. The visualization explicitly reveals how high-discriminability features act as "identity anchors" within clusters, while low-discriminability dimensions function as specific "semantic bridges" between them. This offers a qualitative, "white-box" alternative to projection methods, enabling researchers to diagnose the precise logic governing cluster cohesion and overlap.