A controversial test for Large Language Models concerns the ability to discern possible from impossible language. While some evidence attests to the models' sensitivity to what crosses the limits of grammatically impossible language, this evidence has been contested on the grounds of the soundness of the testing material. We use model-internal representations to tap directly into the way Large Language Models represent the 'grammatical-ungrammatical' distinction. In a novel benchmark, we elicit probabilities from 4 models and compute minimal-pair surprisal differences, juxtaposing probabilities assigned to grammatical sentences to probabilities assigned to (i) lower frequency grammatical sentences, (ii) ungrammatical sentences, (iii) semantically odd sentences, and (iv) pragmatically odd sentences. The prediction is that if string-probabilities can function as proxies for the limits of grammar, the ungrammatical condition will stand out among the conditions that involve linguistic violations, showing a spike in the surprisal rates. Our results do not reveal a unique surprisal signature for ungrammatical prompts, as the semantically and pragmatically odd conditions consistently show higher surprisal. We thus demonstrate that probabilities do not constitute reliable proxies for model-internal representations of syntactic knowledge. Consequently, claims about models being able to distinguish possible from impossible language need verification through a different methodology.

The growing demand for energy-efficient, high-performance AI systems has led to increased attention on alternative computing platforms (e.g., photonic, neuromorphic) due to their potential to accelerate learning and inference. However, integrating such physical components into deep learning pipelines remains challenging, as physical devices often offer limited expressiveness, and their non-differentiable nature renders on-device backpropagation difficult or infeasible. This motivates the development of hybrid architectures that combine digital neural networks with reconfigurable physical layers, which effectively behave as black boxes. In this work, we present a framework for the end-to-end training of such hybrid networks. This framework integrates stochastic zeroth-order optimization for updating the physical layer's internal parameters with a dynamic low-rank surrogate model that enables gradient propagation through the physical layer. A key component of our approach is the implicit projector-splitting integrator algorithm, which updates the lightweight surrogate model after each forward pass with minimal hardware queries, thereby avoiding costly full matrix reconstruction. We demonstrate our method across diverse deep learning tasks, including: computer vision, audio classification, and language modeling. Notably, across all modalities, the proposed approach achieves near-digital baseline accuracy and consistently enables effective end-to-end training of hybrid models incorporating various non-differentiable physical components (spatial light modulators, microring resonators, and Mach-Zehnder interferometers). This work bridges hardware-aware deep learning and gradient-free optimization, thereby offering a practical pathway for integrating non-differentiable physical components into scalable, end-to-end trainable AI systems.

Artificial intelligence has shown significant promise in chest radiography, where deep learning models can approach radiologist-level diagnostic performance. Progress has been accelerated by large public datasets such as MIMIC-CXR, ChestX-ray14, PadChest, and CheXpert, which provide hundreds of thousands of labelled images with pathology annotations. However, these datasets also present important limitations. Automated label extraction from radiology reports introduces errors, particularly in handling uncertainty and negation, and radiologist review frequently disagrees with assigned labels. In addition, domain shift and population bias restrict model generalisability, while evaluation practices often overlook clinically meaningful measures. We conduct a systematic analysis of these challenges, focusing on label quality, dataset bias, and domain shift. Our cross-dataset domain shift evaluation across multiple model architectures revealed substantial external performance degradation, with pronounced reductions in AUPRC and F1 scores relative to internal testing. To assess dataset bias, we trained a source-classification model that distinguished datasets with near-perfect accuracy, and performed subgroup analyses showing reduced performance for minority age and sex groups. Finally, expert review by two board-certified radiologists identified significant disagreement with public dataset labels. Our findings highlight important clinical weaknesses of current benchmarks and emphasise the need for clinician-validated datasets and fairer evaluation frameworks.

The rising computational and energy demands of deep learning, particularly in large-scale architectures such as foundation models and large language models (LLMs), pose significant challenges to sustainability. Traditional gradient-based training methods are inefficient, requiring numerous iterative updates and high power consumption. To address these limitations, we propose a hybrid framework that combines hierarchical decomposition with FPGA-based direct equation solving and incremental learning. Our method divides the neural network into two functional tiers: lower layers are optimized via single-step equation solving on FPGAs for efficient and parallelizable feature extraction, while higher layers employ adaptive incremental learning to support continual updates without full retraining. Building upon this foundation, we introduce the Compound LLM framework, which explicitly deploys LLM modules across both hierarchy levels. The lower-level LLM handles reusable representation learning with minimal energy overhead, while the upper-level LLM performs adaptive decision-making through energy-aware updates. This integrated design enhances scalability, reduces redundant computation, and aligns with the principles of sustainable AI. Theoretical analysis and architectural insights demonstrate that our method reduces computational costs significantly while preserving high model performance, making it well-suited for edge deployment and real-time adaptation in energy-constrained environments.

Air pollution remains a critical threat to public health and environmental sustainability, yet conventional monitoring systems are often constrained by limited spatial coverage and accessibility. This paper proposes an AI-driven agent that predicts ambient air pollution levels from sky images and synthesizes realistic visualizations of pollution scenarios using generative modeling. Our approach combines statistical texture analysis with supervised learning for pollution classification, and leverages vision-language model (VLM)-guided image generation to produce interpretable representations of air quality conditions. The generated visuals simulate varying degrees of pollution, offering a foundation for user-facing interfaces that improve transparency and support informed environmental decision-making. These outputs can be seamlessly integrated into intelligent applications aimed at enhancing sit-uational awareness and encouraging behavioral responses based on real-time forecasts. W e validate our method using a dataset of urban sky images and demonstrate its effectiveness in both pollution level estimation and semantically consistent visual synthesis. The system design further incorporates human-centered user experience principles to ensure accessibility, clarity, and public engagement in air quality forecasting. T o support scalable and energy-efficient deployment, future iterations will incorporate a green CNN architecture enhanced with FPGA-based incremental learning, enabling real-time inference on edge platforms.

We present a novel framework that leverages time series clustering to improve internet traffic matrix (TM) prediction using deep learning (DL) models. Traffic flows within a TM often exhibit diverse temporal behaviors, which can hinder prediction accuracy when training a single model across all flows. To address this, we propose two clustering strategies, source clustering and histogram clustering, that group flows with similar temporal patterns prior to model training. Clustering creates more homogeneous data subsets, enabling models to capture underlying patterns more effectively and generalize better than global prediction approaches that fit a single model to the entire TM. Compared to existing TM prediction methods, our method reduces RMSE by up to 92\% for Abilene and 75\% for GÉANT. In routing scenarios, our clustered predictions also reduce maximum link utilization (MLU) bias by 18\% and 21\%, respectively, demonstrating the practical benefits of clustering when TMs are used for network optimization.

Abstract--This paper develops alternative hyperparameters for specifying sparse Recurrent Neural Networks (RNNs). These hyperparameters allow for varying sparsity within the trainable weight matrices of the model while improving overall performance. This architecture enables the definition of a novel metric, hidden proportion, which seeks to balance the distribution of unknowns within the model and provides significant explanatory power of model performance. T ogether, the use of the varied sparsity RNN architecture combined with the hidden proportion metric generates significant performance gains while improving performance expectations on an a priori basis. This combined approach provides a path forward towards generalized meta-learning applications and model optimization based on intrinsic characteristics of the data set, including input and output dimensions. Selection and specification of neural networks remains an actively studied arena [1]-[3]. Practitioners traditionally follow a pathway of first selecting a model architecture suited to the general characteristics of the data set. Afterwards, model hyperparameters are often optimized through a process involving repeated training runs and cross-validation [4]. The search for optimal hyperparameters is often costly as the total potential combinations grow exponentially with the number of hyperparameters and model effectiveness can often only be judged after training.

This paper addresses the synthetic-to-real domain gap in object detection, focusing on training a YOLOv11 model to detect a specific object (a soup can) using only synthetic data and domain randomization strategies. The methodology involves extensive experimentation with data augmentation, dataset composition, and model scaling. While synthetic validation metrics were consistently high, they proved to be poor predictors of real-world performance. Consequently, models were also evaluated qualitatively, through visual inspection of predictions, and quantitatively, on a manually labeled real-world test set, to guide development. Final mAP@50 scores were provided by the official Kaggle competition. Key findings indicate that increasing synthetic dataset diversity, specifically by including varied perspectives and complex backgrounds, combined with carefully tuned data augmentation, were crucial in bridging the domain gap. The best performing configuration, a YOLOv11l model trained on an expanded and diverse dataset, achieved a final mAP@50 of 0.910 on the competition's hidden test set. This result demonstrates the potential of a synthetic-only training approach while also highlighting the remaining challenges in fully capturing real-world variability.

Directional forecasting in financial markets requires both accuracy and interpretability. Before the advent of deep learning, interpretable approaches based on human-defined patterns were prevalent, but their structural vagueness and scale ambiguity hindered generalization. In contrast, deep learning models can effectively capture complex dynamics, yet often offer limited transparency. To bridge this gap, we propose a two-stage framework that integrates unsupervised pattern extracion with interpretable forecasting. (i) SIMPC segments and clusters multivariate time series, extracting recurrent patterns that are invariant to amplitude scaling and temporal distortion, even under varying window sizes. (ii) JISC-Net is a shapelet-based classifier that uses the initial part of extracted patterns as input and forecasts subsequent partial sequences for short-term directional movement. Experiments on Bitcoin and three S&P 500 equities demonstrate that our method ranks first or second in 11 out of 12 metric--dataset combinations, consistently outperforming baselines. Unlike conventional deep learning models that output buy-or-sell signals without interpretable justification, our approach enables transparent decision-making by revealing the underlying pattern structures that drive predictive outcomes.

This study investigates the use of generative AI to support formative assessment through machine generated reviews of peer reviews in graduate online courses in a public university in the United States. Drawing on Systemic Functional Linguistics and Appraisal Theory, we analyzed 120 metareviews to explore how generative AI feedback constructs meaning across ideational, interpersonal, and textual dimensions. The findings suggest that generative AI can approximate key rhetorical and relational features of effective human feedback, offering directive clarity while also maintaining a supportive stance. The reviews analyzed demonstrated a balance of praise and constructive critique, alignment with rubric expectations, and structured staging that foregrounded student agency. By modeling these qualities, AI metafeedback has the potential to scaffold feedback literacy and enhance leaner engagement with peer review.