Heuristic algorithms play a vital role in solving combinatorial optimization (CO) problems, yet traditional designs depend heavily on manual expertise and struggle to generalize across diverse instances. We introduce \textbf{HeurAgenix}, a two-stage hyper-heuristic framework powered by large language models (LLMs) that first evolves heuristics and then selects among them automatically. In the heuristic evolution phase, HeurAgenix leverages an LLM to compare seed heuristic solutions with higher-quality solutions and extract reusable evolution strategies. During problem solving, it dynamically picks the most promising heuristic for each problem state, guided by the LLM's perception ability. For flexibility, this selector can be either a state-of-the-art LLM or a fine-tuned lightweight model with lower inference cost. To mitigate the scarcity of reliable supervision caused by CO complexity, we fine-tune the lightweight heuristic selector with a dual-reward mechanism that jointly exploits singals from selection preferences and state perception, enabling robust selection under noisy annotations. Extensive experiments on canonical benchmarks show that HeurAgenix not only outperforms existing LLM-based hyper-heuristics but also matches or exceeds specialized solvers. Code is available at https://github.com/microsoft/HeurAgenix.

Increasing resolution and coverage of astrophysical and climate data necessitates increasingly sophisticated models, often pushing the limits of computational feasibility. While emulation methods can reduce calculation costs, the neural architectures typically used--optimised via gradient descent--are themselves computationally expensive to train, particularly in terms of data generation requirements. This paper investigates the utility of the Extreme Learning Machine (ELM) as a lightweight, non-gradient-based machine learning algorithm for accelerating complex physical models. We evaluate ELM surrogate models in two test cases with different data structures: (i) sequentially-structured data, and (ii) image-structured data. For test case (i), where the number of samples $N$ >> the dimensionality of input data $d$, ELMs achieve remarkable efficiency, offering a 100,000$\times$ faster training time and a 40$\times$ faster prediction speed compared to a Bi-Directional Recurrent Neural Network (BIRNN), whilst improving upon BIRNN test performance. For test case (ii), characterised by $d >> N$ and image-based inputs, a single ELM was insufficient, but an ensemble of 50 individual ELM predictors achieves comparable accuracy to a benchmark Convolutional Neural Network (CNN), with a 16.4$\times$ reduction in training time, though costing a 6.9$\times$ increase in prediction time. We find different sample efficiency characteristics between the test cases: in test case (i) individual ELMs demonstrate superior sample efficiency, requiring only 0.28% of the training dataset compared to the benchmark BIRNN, while in test case (ii) the ensemble approach requires 78% of the data used by the CNN to achieve comparable results--representing a trade-off between sample efficiency and model complexity.

Modern tokenizers employ deterministic algorithms to map text into a single "canonical" token sequence, yet the same string can be encoded as many non-canonical tokenizations using the tokenizer vocabulary. In this work, we investigate the robustness of LMs to text encoded with non-canonical tokenizations entirely unseen during training. Surprisingly, when evaluated across 20 benchmarks, we find that instruction-tuned models retain up to 93.4% of their original performance when given a randomly sampled tokenization, and 90.8% with character-level tokenization. We see that overall stronger models tend to be more robust, and robustness diminishes as the tokenization departs farther from the canonical form. Motivated by these results, we then identify settings where non-canonical tokenization schemes can *improve* performance, finding that character-level segmentation improves string manipulation and code understanding tasks by up to +14%, and right-aligned digit grouping enhances large-number arithmetic by +33%. Finally, we investigate the source of this robustness, finding that it arises in the instruction-tuning phase. We show that while both base and post-trained models grasp the semantics of non-canonical tokenizations (perceiving them as containing misspellings), base models try to mimic the imagined mistakes and degenerate into nonsensical output, while post-trained models are committed to fluent responses. Overall, our findings suggest that models are less tied to their tokenizer than previously believed, and demonstrate the promise of intervening on tokenization at inference time to boost performance.

E-commerce information retrieval (IR) systems struggle to simultaneously achieve high accuracy in interpreting complex user queries and maintain efficient processing of vast product catalogs. The dual challenge lies in precisely matching user intent with relevant products while managing the computational demands of real-time search across massive inventories. In this paper, we propose a Nested Embedding Approach to product Retrieval and Ranking, called NEAR$^2$, which can achieve up to $12$ times efficiency in embedding size at inference time while introducing no extra cost in training and improving performance in accuracy for various encoder-based Transformer models. We validate our approach using different loss functions for the retrieval and ranking task, including multiple negative ranking loss and online contrastive loss, on four different test sets with various IR challenges such as short and implicit queries. Our approach achieves an improved performance over a smaller embedding dimension, compared to any existing models.

We present an enhanced YOLOv8 real time vehicle detection and classification framework, for estimating carbon emissions in urban environments. The system enhances YOLOv8 architecture to detect, segment, and track vehicles from live traffic video streams. Once a vehicle is localized, a dedicated deep learning-based identification module is employed to recognize license plates and classify vehicle types. Since YOLOv8 lacks the built-in capacity for fine grained recognition tasks such as reading license plates or determining vehicle attributes beyond class labels, our framework incorporates a hybrid pipeline where each detected vehicle is tracked and its bounding box is cropped and passed to a deep Optical Character Recognition (OCR) module. This OCR system, composed of multiple convolutional neural network (CNN) layers, is trained specifically for character-level detection and license plate decoding under varied conditions such as motion blur, occlusion, and diverse font styles. Additionally, the recognized plate information is validated using a real time API that cross references with an external vehicle registration database to ensure accurate classification and emission estimation. This multi-stage approach enables precise, automated calculation of per vehicle carbon emissions. Extensive evaluation was conducted using a diverse vehicle dataset enriched with segmentation masks and annotated license plates. The YOLOv8 detector achieved a mean Average Precision (mAP@0.5) of approximately 71% for bounding boxes and 70% for segmentation masks. Character level OCR accuracy reached up to 99% with the best performing CNN model. These results affirm the feasibility of combining real time object detection with deep OCR for practical deployment in smart transportation systems, offering a scalable solution for automated, vehicle specific carbon emission monitoring.

The pursuit of efficient and controllable high-quality content generation remains a central challenge in artificial intelligence-generated content (AIGC). While one-step generators, enabled by diffusion distillation techniques, offer excellent generation quality and computational efficiency, adapting them to new control conditions--such as structural constraints, semantic guidelines, or external inputs--poses a significant challenge. Conventional approaches often necessitate computationally expensive modifications to the base model and subsequent diffusion distillation. This paper introduces Noise Consistency Training (NCT), a novel and lightweight approach to directly integrate new control signals into pre-trained one-step generators without requiring access to original training images or retraining the base diffusion model. NCT operates by introducing an adapter module and employs a noise consistency loss in the noise space of the generator. This loss aligns the adapted model's generation behavior across noises that are conditionally dependent to varying degrees, implicitly guiding it to adhere to the new control. Theoretically, this training objective can be understood as minimizing the distributional distance between the adapted generator and the conditional distribution induced by the new conditions. NCT is modular, data-efficient, and easily deployable, relying only on the pre-trained one-step generator and a control signal model. Extensive experiments demonstrate that NCT achieves state-of-the-art controllable generation in a single forward pass, surpassing existing multi-step and distillation-based methods in both generation quality and computational efficiency. Code is available at https://github.com/Luo-Yihong/NCT

Matrix seriation, the problem of permuting the rows and columns of a matrix to uncover latent structure, is a fundamental technique in data science, particularly in the visualization and analysis of relational data. Applications span clustering, anomaly detection, and beyond. In this work, we present a unified framework grounded in mathematical optimization to address matrix seriation from a rigorous, model-based perspective. Our approach leverages combinatorial and mixed-integer optimization to represent seriation objectives and constraints with high fidelity, bridging the gap between traditional heuristic methods and exact solution techniques. We introduce new mathematical programming models for neighborhood-based stress criteria, including nonlinear formulations and their linearized counterparts. For structured settings such as Moore and von Neumann neighborhoods, we develop a novel Hamiltonian path-based reformulation that enables effective control over spatial arrangement and interpretability in the reordered matrix. To assess the practical impact of our models, we carry out an extensive set of experiments on synthetic and real-world datasets, as well as on a newly curated benchmark based on a coauthorship network from the matrix seriation literature. Our results show that these optimization-based formulations not only enhance solution quality and interpretability but also provide a versatile foundation for extending matrix seriation to new domains in data science.

Prior work on the learnability of transformers has established its capacity to approximate specific algorithmic patterns through training under restrictive architectural assumptions. Fundamentally, these arguments remain data-driven and therefore can only provide a probabilistic guarantee. Expressivity, on the contrary, has theoretically been explored to address the problems \emph{computable} by such architecture. These results proved the Turing-completeness of transformers, investigated bounds focused on circuit complexity, and formal logic. Being at the crossroad between learnability and expressivity, the question remains: \emph{can transformer architectures exactly simulate an arbitrary attention mechanism, or in particular, the underlying operations?} In this study, we investigate the transformer encoder's ability to simulate a vanilla attention mechanism. By constructing a universal simulator $\mathcal{U}$ composed of transformer encoders, we present algorithmic solutions to identically replicate attention outputs and the underlying elementary matrix and activation operations via RASP, a formal framework for transformer computation. Our proofs, for the first time, show the existence of an algorithmically achievable data-agnostic solution, previously known to be approximated only by learning.

Anatomical trees are critical for clinical diagnosis and treatment planning, yet their complex and diverse geometry make accurate representation a significant challenge. Motivated by the latest advances in large language models, we introduce an autoregressive method for synthesizing anatomical trees. Our approach first embeds vessel structures into a learned discrete vocabulary using a VQ-VAE architecture, then models their generation autoregressively with a GPT-2 model. This method effectively captures intricate geometries and branching patterns, enabling realistic vascular tree synthesis. Comprehensive qualitative and quantitative evaluations reveal that our technique achieves high-fidelity tree reconstruction with compact discrete representations. Moreover, our B-spline representation of vessel cross-sections preserves critical morphological details that are often overlooked in previous' methods parameterizations. To the best of our knowledge, this work is the first to generate blood vessels in an autoregressive manner. Code is available at https://github.com/LIA-DiTella/VesselGPT-MICCAI.

Four killed in Kyiv in new Russian aerial attack 12 minutes agoShareSaveJaroslav LukivBBC NewsShareSaveUkraine's emergencies service DSNSRescuers from Ukraine's emergencies service DSNS tackle fire in a residential building destroyed in the latest Russian attack on Kyiv At least four people have been killed in an overnight Russian missile and drone attack on Ukraine's capital Kyiv, the interior minister says. In a post on social media, Ihor Klymenko says residential areas, hospitals and sports infrastructure were hit. "An entire section of a residential high-rise building was destroyed" in the worst-hit Shevchenkivskyi district, he says, adding that some people are trapped under the rubble. In the Kyiv region, a woman was killed and another two people injured in the Russian aerial attack, regional head Mykola Kalashnyk says. The Russian military has not commented on the issue.