Abstract

Unreinforced masonry (URM) buildings constitute a substantial share of the global building stock
and remain highly vulnerable to seismic loading owing to their brittle response and negligible ten­
sile capacity. Flexural and shear failures are the most prevalent in-plane mechanisms governing
deformation capacity and collapse potential, making reliable identification of the governing failure
mode from crack patterns essential for damage assessment and retrofit decisions. However, inter­
pretation of cracking morphology remains inconsistent in practice, since it requires linkage to the
governing stress state and contextual parameters rather than appearance alone. This review syn­
thesizes the mechanics of flexural and shear crack formation in URM walls under in-plane loading,
as influenced by axial load ratio, aspect ratio, boundary conditions, and masonry anisotropy, and
examines transitional and mixed-mode responses that often cause post-earthquake misclassifica­
tion. A mechanics-based diagnostic framework is then proposed for crack-based identification of
failure mechanisms, operationalized as a sequential decision tree linking crack morphology, spatial
distribution, deformation kinematics, and governing parameters. A quantitative crack dominance
ratio (CDR) is proposed as a practical indicator for mixed-mode classification using experimentally
validated thresholds. The framework is illustrated against three experimental cases, producing clas­
sifications consistent with reported failure modes, and key research gaps are identified.

Abstract

Offline safe reinforcement learning learns high-return policies that satisfy hard safety constraints using only a pre-collected dataset. This setting is challenging due to the inability to explore, and the risk of propagating value errors through unsafe state-space regions. To address this, first, we characterize the safe state region by developing a framework for learning control barrier functions (CBFs) using a novel generalized Bellman operator, yielding a persistent safety set, from which the agent can remain safe indefinitely. Second, we show that several existing safety set estimation methods (e.g., reachability-constrained RL) can be formulated within our CBF learning framework, highlighting its generality. We further propose a new CBF that ensures safety under environment dynamics uncertainty, unlike standard CBFs designed for deterministic settings. Third, we propose a new reward maximization algorithm that effectively exploits our learned persistent safety set for reward critic estimation. Empirical results on standard benchmarks show that our approach achieves state-of-the-art safety with fewer constraint violations while maintaining competitive returns.

Abstract

Sparse-accelerator design spaces are usually searched against analytical models, so a design point is admitted on what a model predicts rather than on what the hardware does. SparseCraft closes that gap with a language model inside a closed CHIA loop. In each of 15 iterations the model reads the measured outcome of the previous one and edits the Chisel RTL, the memory configuration and the sparse-kernel schedule of a Gemmini accelerator through MCP tool servers, and no candidate counts until it has been checked for legality, elaborated, simulated cycle-accurately, checked bit-for-bit on every output against a golden reference, and synthesised. The harness turns each measurement into the next work order, a diagnosed bottleneck with matching strategy guidance, the history of tried designs and a score of the model's own prediction, and a second model repairs changes that fail a gate. On a 512×512 GraphChallenge sparse-DNN layer the loop reaches 2.1x fewer cycles, 9.8x less off-chip traffic and 22.8% less area than the block-sparse Gemmini baseline, with 5.61x higher modelled perf/W and 11.8x lower EDP. The levers span three layers: a schedule that keeps the dense operand resident removes 9.8x of the traffic, a zero-gated MAC and a zero-row skip unit that the model wrote in Chisel cut energy, and resizing the memories cuts area.

Abstract

Given a single image of a room and a sentence such as ``a person walks to the black armchair and sits down,'' producing a 3D motion sequence of a virtual person performing the described action is out of reach for existing methods. Scene-conditioned methods need hours of paired (scene, motion, language) capture. Zero-shot methods need a pre-scanned 3D scene. Image-based methods produce only static poses. We propose a training-free pipeline that synthesizes scene-aware human-scene interaction (HSI) motion from a single RGB image and a free-form text prompt, with no HSI-specific training and no pre-scanned 3D scene. The key observation is that every needed capability already exists as a frozen foundation model: the bottleneck is composition, not capability. Our method, ENACT, uses a vision-language model (VLM) for high-level reasoning. The VLM parses the input image and prompt into one structured keyframe per sentence, listing the action verb, the body part, the target object, a six-bin contact yaw, an image-edit prompt, and a foot-grounding flag. The keyframe then drives six frozen specialists: a monocular 3D model for the scene point cloud, an open-vocabulary segmenter for object localization, a single-view mesh reconstructor for object geometry, an instruction-following image editor for a synthetic interaction image, an image-conditioned human-mesh recoverer for pose initialization, and a 2D-grounded 3D contact predictor for object affordance. Per-keyframe contact poses are optimized under a whole-body diffusion prior, and a constraint-conditioned motion diffuser connects successive keyframes into one continuous motion. ENACT achieves the lowest mean and maximum scene penetration and the lowest foot sliding among all baselines on a standard HSI benchmark, despite consuming the smallest input. ENACT also generalizes to synthetic datasets and to in-the-wild phone images, including chained multi-step interactions.

Abstract

Digital platforms increasingly influence the visibility and emotional dissemination of political messages; however, their ability to convert engagement into electoral influence is uneven and contingent upon context. Drawing on computational analysis of Facebook and Instagram during the 2024 Andhra Pradesh Assembly elections, it analyses how platform-specific features, emotive content, and temporal patterns influence voter engagement and political prominence. The paper looks at three important aspects of digital campaigning. First, it looks at posting patterns and campaign intensity, focusing on how frequently political actor's post. Second, it looks how different types of posts, such as those about welfare, leadership, or direct political attacks generate higher levels of interaction among voters. Third, it studies how the campaign narratives change over time, rise and fall of key themes in response to political events and moments.
This study analyzes over 30,000 social media posts posted between June 2023 and May 2024 by the two principal political parties in Andhra Pradesh, i.e., the Telugu Desam Party(TDP) and the Yuvajana Sramika Rythu Congress Party(YSRCP), along with their respective leaders, N. Chandrababu Naidu and Y. S. Jagan Mohan Reddy. After collecting the raw data, the raw data sets were sorted chronologically and split by predefined timelines. The time frame has been divided into three timelines or campaign phases, considering the various political endpoints. Engagement rate was calculated using standardized measures, and posts were filtered to exclude non-political content through a rule-based classification process, allowing the analysis to focus exclusively on substantive political communication. For a better understanding of the voters' perspective, we also conducted sentiment and thematic analysis of the most relevant comments. This allows us to study how voters interpret the campaign messages.
The results indicate that digital campaigning is predominantly centered around particular political events, announcements, controversies, and campaign milestones. Posts with emotional content and personal attacking of the leaders have a high engagement rate. It was also evident that Instagram, which is visual-centric, has a much higher engagement rate than Facebook. Facebook supported longer political messaging and more sustained interaction in comment sections.
By using computational tools to analyze engagement dynamics, the study enables the field of electoral politics to move beyond descriptive explanations of digital campaigning to more detailed examinations of political communication and digitally structured participation. This approach opens new avenues for research on electoral politics by demonstrating that digital engagement can be systematically studied as a political process.

Abstract

Neural Style Transfer (NST) traditionally applies a fixed style to a target image, offering limited control over its degree, progression, or variations. Recent methods have introduced various control mechanisms like spatial attention, entropy regularization, and latent space manipulation. However, these approaches lack interpretable, human-understandable controls for fine-grained stylistic control as intermediate stylization is typically achieved by adjusting content and style losses. Other works explore interpolation across feature layers, but they operate through abstract parameters that require domain expertise and often lack intuitive control, as the relationship between parameters and stylization remains unclear due to the black-box nature of the architecture. In this paper, we introduce CAV Styler, a human-interpretable style transfer approach. We exploit human interpretable concepts such as textures, blurs, and style details to enable fine-grained control over stylization. To our knowledge, ours is among the first in image-to-image style transfer to offer explicit user guidance through semantically meaningful parameters. Our method demonstrates superior controllability compared to existing approaches while maintaining competitive stylization quality, opening new possibilities for intuitive and precise artistic style manipulation.

Abstract

Existing translation models are typically trained on sentence-level and formal text, limiting their ability to capture everyday conversational dialogue phenomena such as informality, speaker interaction, and discourse
coherence.
Most existing Indic translation resources and evaluation benchmarks focus on
sentence-level or formal text, making it difficult to assess translation quality of the dialogue phenomena.
In this work, we introduce textit{BaatCheet}, a multilingual dialogue corpus named after
the Hindi term for textit{conversation} or textit{chitchat},
containing approximately 49,000 dialogues for dialogue translation across five
translation directions.
We fine-tune five open-source LLMs across seven training data configurations
and find that fine-tuning yields substantial gains over zero- and few-shot
baselines.
To comprehensively evaluate dialogue translation quality,
we employ multiple evaluation strategies, including automatic metrics, LLM-as-judge, and human assessments using an SQM-guided Direct Assessment (DA) Protocol.

Abstract

Gaussian Processes (GPs) are widely used as surrogates for black-box functions in sequential decision-making problems such as Bayesian optimization (BO), level set estimation (LSE), and Bayesian active learning (BAL). GP performance critically depends on kernels, and standard kernels can lead to suboptimal decisions under misspecification. To address this, we introduce HACK GPs (Hedge Adaptive Cumulative Kernels), a method that views kernel selection as an online learning with expert advice problem. HACK treats each candidate kernel as a GP "expert" and updates a distribution over experts online using AdaHedge, based on a loss received as a proxy for their ability to fit the function and align with the task objective. We provide two variants of HACK: (i) Mixture of Gaussians (MoG) and (ii) categorical sampling. We establish general guarantees showing that, under a loss-gap condition, the weight concentrates on the best kernel and the resulting acquisition function is close to that of the best expert. Empirically, we observe robust performance across BO, LSE, and BAL compared to standard kernels such as Squared Exponential and Matern-5/2, as well as simple ensemble baselines.

Abstract

A fundamental tension in coded Private Information Retrieval (PIR) lies in balancing the storage overhead placed on servers against the communication and query complexity demanded from the user. The Woodruff-Yekhanin (WY) scheme~cite{WY05} and cite{DGY11, DG15, GKS25} achieve low query complexity by letting servers compute high-order directional derivatives over a fully replicated Reed-Muller codeword, while the Lavauzelle-Nardi (LN) scheme~cite{LN21} minimizes storage to a single encoded curve per server, at the price of maximum query complexity. We establish that a genuine intermediate multiplicity-storage tradeoff exists emph{if and only if} the line-degree parameter $omega geq 2$. For $omega = 1$ (affine-line-based coding), enabling even first-order multiplicity forces complete WY-like codeword replication. For Weighted Reed-Muller (WRM) codes with weight $eta = omega geq 2$, the non-surjectivity of polynomials of degree at least 2 over finite fields creates algebraic "slack" that supports fractional co-location: a server stores $t = s+1$ parallel-class curves, and a Vandermonde inversion lets the user extract the desired derivative while keeping the private query direction hidden. Conditional on a natural Polynomial Image Independence Conjecture (PIIC), whose $s=1$ case is the classical Weil bound, the marginal storage cost of each additional multiplicity unit follows a geometric series that halves at every step. The protocol achieves strict $omega$-privacy: any coalition of up to $omega$ servers learns nothing about the target index.

Abstract

In the modern era, metasurfaces have gained significant attention for their potential in light manipulation, including focusing, bending, blocking, and wavefront control, as well as designing ultra-thin optical components. Metasurfaces comprising highindex dielectric nanostructures have emerged as an inevitable platform for high-performance flat optical devices. Among these, a dichroic beam splitter is a crucial optical component that selectively reflects and transmits light based on the wavelength, enabling precise control over beam routing. In this work, we demonstrate a polarization- and angle-tolerant metasurface-based dichroic beam splitter that operates as a harmonic separator, efficiently separating the fundamental wavelength (1560 nm) and its second harmonic (780 nm) wavelength. Using a 2D array of gallium phosphide (GaP) cylindrical nanostructures, the proposed device exhibits high transmission at 1560 nm while simultaneously achieving near-unity reflection at 780 nm over a bandwidth of approximately 168 nm. This device finds application in fluorescence microscopy, multispectral imaging and advanced laser systems due to its precise spectral selectivity in the near-infrared regime. Beyond these applications, our result highlights a versatile design route to compact dichroic beam splitters that can replace conventional bulky optical devices, with promising prospects for on-chip photonics and portable optical elements.