Large Language Models (LLMs) are increasingly deployed in mission-critical systems, facilitating tasks such as satellite operations, command-and-control, military decision support, and cyber defense. Many of these systems are accessed through application programming interfaces (APIs). When such APIs lack robust access controls, they can expose full or top-k logits, creating a significant and often overlooked attack surface. Prior art has mainly focused on reconstructing the output projection layer or distilling surface-level behaviors. However, regenerating a black-box model under tight query constraints remains underexplored. We address that gap by introducing a constrained replication pipeline that transforms partial logit leakage into a functional deployable substitute model clone. Our two-stage approach (i) reconstructs the output projection matrix by collecting top-k logits from under 10k black-box queries via singular value decomposition (SVD) over the logits, then (ii) distills the remaining architecture into compact student models with varying transformer depths, trained on an open source dataset. A 6-layer student recreates 97.6% of the 6-layer teacher model's hidden-state geometry, with only a 7.31% perplexity increase, and a 7.58 Negative Log-Likelihood (NLL). A 4-layer variant achieves 17.1% faster inference and 18.1% parameter reduction with comparable performance. The entire attack completes in under 24 graphics processing unit (GPU) hours and avoids triggering API rate-limit defenses. These results demonstrate how quickly a cost-limited adversary can clone an LLM, underscoring the urgent need for hardened inference APIs and secure on-premise defense deployments.

Large Language Models (LLMs) are increasingly used for accessibility guidance, yet many disability groups remain underserved by their advice. To address this gap, we present taxonomy aligned benchmark1 of human validated, general purpose accessibility questions, designed to systematically audit inclusivity across disabilities. Our benchmark evaluates models along three dimensions: Question-Level Coverage (breadth within answers), Disability-Level Coverage (balance across nine disability categories), and Depth (specificity of support). Applying this framework to 17 proprietary and open-weight models reveals persistent inclusivity gaps: Vision, Hearing, and Mobility are frequently addressed, while Speech, Genetic/Developmental, Sensory-Cognitive, and Mental Health remain under served. Depth is similarly concentrated in a few categories but sparse elsewhere. These findings reveal who gets left behind in current LLM accessibility guidance and highlight actionable levers: taxonomy-aware prompting/training and evaluations that jointly audit breadth, balance, and depth.

The making of knowledge engines in natural language processing has been shaped by two seemingly distinct paradigms: one grounded in structure, the other driven by massively available unstructured data. The structured paradigm leverages predefined symbolic interactions, such as knowledge graphs, as priors and designs models to capture them. In contrast, the unstructured paradigm centers on scaling transformer architectures with increasingly vast data and model sizes, as seen in modern large language models. Despite their divergence, this thesis seeks to establish conceptual connections bridging these paradigms. Two complementary forces, structure and destructure, emerge across both paradigms: structure organizes seen symbolic interactions, while destructure, through periodic embedding resets, improves model plasticity and generalization to unseen scenarios. These connections form a new recipe for developing general knowledge engines that can support transparent, controllable, and adaptable intelligent systems.

Cities today generate enormous streams of data from sensors, cameras, and connected infrastructure. While this information offers unprecedented opportunities to improve urban life, most existing systems struggle with scale, latency, and fragmented insights. This work introduces a framework that blends physics-informed machine learning, multimodal data fusion, and knowledge graph representation with adaptive, rule-based intelligence powered by large language models (LLMs). Physics-informed methods ground learning in real-world constraints, ensuring predictions remain meaningful and consistent with physical dynamics. Knowledge graphs act as the semantic backbone, integrating heterogeneous sensor data into a connected, queryable structure. At the edge, LLMs generate context-aware rules that adapt filtering and decision-making in real time, enabling efficient operation even under constrained resources. Together, these elements form a foundation for digital twin systems that go beyond passive monitoring to provide actionable insights. By uniting physics-based reasoning, semantic data fusion, and adaptive rule generation, this approach opens new possibilities for creating responsive, trustworthy, and sustainable smart infrastructures.

Transformers achieve state-of-the-art results across many tasks, but their uniform application of quadratic self-attention to every token at every layer makes them computationally expensive. We introduce DTRNet (Dynamic Token Routing Network), an improved Transformer architecture that allows tokens to dynamically skip the quadratic cost of cross-token mixing while still receiving lightweight linear updates. By preserving the MLP module and reducing the attention cost for most tokens to linear, DTRNet ensures that every token is explicitly updated while significantly lowering overall computation. This design offers an efficient and effective alternative to standard dense attention. Once trained, DTRNet blocks routes only ~10% of tokens through attention at each layer while maintaining performance comparable to a full Transformer. It consistently outperforms routing-based layer skipping methods such as MoD and D-LLM in both accuracy and memory at matched FLOPs, while routing fewer tokens to full attention. Its efficiency gains, scales with sequence length, offering significant reduction in FLOPs for long-context inputs. By decoupling token updates from attention mixing, DTRNet substantially reduces the quadratic share of computation, providing a simple, efficient, and scalable alternative to Transformers.

The success of the generative model has gained unprecedented attention in the music generation area. Transformer-based architectures have set new benchmarks for model performance. However, their practical adoption is hindered by some critical challenges: the demand for massive computational resources and inference time, due to their large number of parameters. These obstacles make them infeasible to deploy on edge devices, such as smartphones and wearables, with limited computational resources. In this work, we present TinyMusician, a lightweight music generation model distilled from MusicGen (a State-of-the-art music generation model). The experimental results demonstrate that TinyMusician retains 93% of the MusicGen-Small performance with 55% less model size. TinyMusician is the first mobile-deployable music generation model that eliminates cloud dependency while maintaining high audio fidelity and efficient resource usage. Music, reflecting culture, social classes, ethnic identities, and historical eras, has woven itself into humanity's shared heritage through centuries of evolution (Toynbee, 2012).

Recent developments in applied mathematics increasingly employ machine learning (ML)-particularly supervised learning-to accelerate numerical computations, such as solving nonlinear partial differential equations. In this work, we extend such techniques to objects of a more theoretical nature: the classification and structural analysis of fractal sets. Focusing on the Mandelbrot and Julia sets as principal examples, we demonstrate that supervised learning methods-including Classification and Regression Trees (CART), K-Nearest Neighbors (KNN), Multilayer Perceptrons (MLP), and Recurrent Neural Networks using both Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM), Random Forests (RF), and Convolutional Neural Networks (CNN)-can classify fractal points with significantly higher predictive accuracy and substantially lower computational cost than traditional numerical approaches, such as the thresholding technique. These improvements are consistent across a range of models and evaluation metrics. Notably, KNN and RF exhibit the best overall performance, and comparative analyses between models (e.g., KNN vs. LSTM) suggest the presence of novel regularity properties in these mathematical structures. Collectively, our findings indicate that ML not only enhances classification efficiency but also offers promising avenues for generating new insights, intuitions, and conjectures within pure mathematics.

As early detection of breast cancer strongly favors successful therapeutic outcomes, there is major commercial interest in optimizing breast cancer screening. However, current risk prediction models achieve modest performance and do not incorporate digital breast tomosynthesis (DBT) imaging, which was FDA-approved for breast cancer screening in 2011. To address this unmet need, we present a deep learning (DL)-based framework capable of forecasting an individual patient's 5-year breast cancer risk directly from screening DBT. Using an unparalleled dataset of 161,753 DBT examinations from 50,590 patients, we trained a risk predictor based on features extracted using the Meta AI DINOv2 image encoder, combined with a cumulative hazard layer, to assess a patient's likelihood of developing breast cancer over five years. On a held-out test set, our best-performing model achieved an AUROC of 0.80 on predictions within 5 years. These findings reveal the high potential of DBT-based DL approaches to complement traditional risk assessment tools, and serve as a promising basis for additional investigation to validate and enhance our work.

Recent AI-based scoliosis screening methods primarily rely on large-scale silhouette datasets, often neglecting clinically relevant postural asymmetries--key indicators in traditional screening. In contrast, pose data provide an intuitive skeletal representation, enhancing clinical interpretability across various medical applications. However, pose-based scoliosis screening remains under-explored due to two main challenges: (1) the scarcity of large-scale, annotated pose datasets; and (2) the discrete and noise-sensitive nature of raw pose coordinates, which hinders the modeling of subtle asymmetries. To address these limitations, we introduce Scoliosis1K-Pose, a 2D human pose annotation set that extends the original Scoliosis1K dataset, comprising 447,900 frames of 2D keypoints from 1,050 adolescents. Building on this dataset, we introduce the Dual Representation Framework (DRF), which integrates a continuous skeleton map to preserve spatial structure with a discrete Postural Asymmetry V ec-tor (P A V)that encodes clinically relevant asymmetry descriptors. A novel P A V-Guided Attention (PGA) module further uses the P A V as clinical prior to direct feature extraction from the skeleton map, focusing on clinically meaningful asymmetries. Extensive experiments demonstrate that DRF achieves state-of-the-art performance. Visualizations further confirm that the model leverages clinical asymmetry cues to guide feature extraction and promote synergy between its dual representations.

Large language models (LLMs) have demonstrated exceptional proficiency in language understanding. However, when LLMs align their outputs with deceptive and / or misleading prompts, the generated responses could deviate from the de facto information. Such observations are known as fawning hallucinations, where the model prioritizes alignment with the input's implied perspective over accuracy and truthfulness. In this work, we analyze fawning hallucinations in various natural language processing tasks and tailor the so-termed contrastive decoding method for fawning-hallucination mitigation. Specifically, we design two paradigms to generate corresponding deceptive and / or misleading inputs for the consistent fawning hallucinations induction. Then, we propose the collaborative contrastive decoding (CCD) to handle the fawning hallucinations across di ff erent tasks in LLMs. By contrasting the deviation in output distribution between induced and transformed neutral inputs, the proposed CCD can reduce reliance on deceptive and / or misleading information without requiring additional training. Extensive experiments demonstrate that the proposed CCD can e ff ectively mitigate fawning hallucinations and improve the factuality of the generated responses over various tasks. Introduction Large language models (LLMs), exemplified by the Chat-GPT series [1], have demonstrated their remarkable capabilities across a wide range of natural language processing (NLP) tasks. These tasks include text translation [2, 3], summarization [4, 5], and a ffective computing [6, 7, 8, 9], showcasing the versatility and impact of artificial intelligence (AI). Despite the impressive performance, LLMs are criticized for the potential to generate fabricated, inaccurate, or incorrect information. This phenomenon, known as "hallucination", hinders the further practical application of LLMs.