RBF kernel to increase pretraining data diversity. Architectural details In all experiments, we use the same ExPT architecture. This section details how we constructed new objectives from the original D'Kitty and Ant that we In Ant-Energy, the reward at each time step is: R =1+ Survival reward Control cost Contact cost, (6) which means we incentivize the robot to conserve energy instead of running fast. D'Kitty tasks In D'Kitty, the goal is to design a morphology that allows the D'Kitty robot to reach We found the approximate oracle provided by Design-Bench not accurate enough to provide a reliable comparison of optimization methods on this task. C.1 Effects of GP hyperparameters We empirically examine the impact of two GP hyperparameters, the variance and the length scale ` Specifically, we evaluate the performance of ExPT on D'Kitty We average the performance across 3 seeds.

For each dataset, all unprocessed raw files are represented in .json The datasets are subject to the MIT license. In this subsection, we further analyze the link prediction from the various models applied in the study. Table 6 and 7 represent the effect of link prediction on different datasets from various distinct. In this subsection, we further analyze the node classification results from various models.

Transformers underpin modern large language models (LLMs) and are commonly assumed to be behaviorally unstructured at random initialization, with all meaningful preferences emerging only through large-scale training. We challenge this assumption by showing that randomly initialized transformers already exhibit strong and systematic structural biases. In particular, untrained models display extreme token preferences: across random input sequences, certain tokens are predicted with probabilities orders of magnitude larger. We provide a mechanistic explanation for this phenomenon by dissecting the transformer architecture at initialization. We show that extreme token preference arises from a contraction of token representations along a random seed-dependent direction. This contraction is driven by two interacting forces: (i) asymmetric nonlinear activations in MLP sublayers induce global (inter-sequence) representation concentration, and (ii) self-attention further amplifies this effect through local (intra-sequence) aggregation. Together, these mechanisms align hidden representations along a direction determined solely by the random initialization, producing highly non-uniform next-token predictions. Beyond mechanistic insight, we demonstrate that these initialization-induced biases persist throughout training, forming a stable and intrinsic model identity. Leveraging this property, we introduce SeedPrint, a fingerprinting method that can reliably distinguish models that differ only in their random initialization, even after extensive training and under substantial distribution shift. Finally, we identify a fundamental positional discrepancy inherent to the attention mechanism's intra-sequence contraction that is causally linked to the attention-sink phenomenon. This discovery provides a principled explanation for the emergence of sinks and offers a pathway for their control.

Early detection and characterization of coronavirus disease (COVID-19), caused by SARS-CoV-2, remain critical for effective clinical response and public-health planning. The global availability of large-scale viral sequence data presents significant opportunities for computational analysis; however, existing approaches face notable limitations. Phylogenetic tree-based methods are computationally intensive and do not scale efficiently to today's multi-million-sequence datasets. Similarly, current embedding-based techniques often rely on aligned sequences or exhibit suboptimal predictive performance and high runtime costs, creating barriers to practical large-scale analysis. In this study, we focus on the most prevalent SARS-CoV-2 lineages associated with the spike protein region and introduce a scalable embedding method that leverages hashing to generate compact, low-dimensional representations of spike sequences. These embeddings are subsequently used to train a variety of machine learning models for supervised lineage classification. We conduct an extensive evaluation comparing our approach with multiple baseline and state-of-the-art biological sequence embedding methods across diverse metrics. Our results demonstrate that the proposed embeddings offer substantial improvements in efficiency, achieving up to 86.4\% classification accuracy while reducing embedding generation time by as much as 99.81\%. This highlights the method's potential as a fast, effective, and scalable solution for large-scale viral sequence analysis.

Drug-drug interactions (DDIs) remain a major source of preventable harm, and many clinically important mechanisms are still unknown. Existing models either rely on pharmacologic knowledge graphs (KGs), which fail on unseen drugs, or on electronic health records (EHRs), which are noisy, temporal, and site-dependent. We introduce, to our knowledge, the first system that conditions KG relation scoring on patient-level EHR context and distills that reasoning into an EHR-only model for zero-shot inference. A fusion "Teacher" learns mechanism-specific relations for drug pairs represented in both sources, while a distilled "Student" generalizes to new or rarely used drugs without KG access at inference. Both operate under a shared ontology (set) of pharmacologic mechanisms (drug relations) to produce interpretable, auditable alerts rather than opaque risk scores. Trained on a multi-institution EHR corpus paired with a curated DrugBank DDI graph, and evaluated using a a clinically aligned, decision-focused protocol with leakage-safe negatives that avoid artificially easy pairs, the system maintains precision across multi-institutuion test data, produces mechanism-specific, clinically consistent predictions, reduces false alerts (higher precision) at comparable overall detection performance (F1), and misses fewer true interactions compared to prior methods. Case studies further show zero-shot identification of clinically recognized CYP-mediated and pharmacodynamic mechanisms for drugs absent from the KG, supporting real-world use in clinical decision support and pharmacovigilance.

Scaling language models unlocks impressive capabilities, but the accompanying computational and memory demands make both training and deployment expensive. Existing efficiency efforts typically target either parameter sharing or adaptive computation, leaving open the question of how to attain both simultaneously. We introduce Mixture-of-Recursions (MoR), a unified framework that combines the two axes of efficiency inside a single Recursive Transformer. MoR reuses a shared stack of layers across recursion steps to achieve parameter efficiency, while lightweight routers enable adaptive token-level thinking by dynamically assigning different recursion depths to individual tokens. This allows MoR to focus quadratic attention computation only among tokens still active at a given recursion depth, further improving memory access efficiency by selectively caching only their key-value pairs. Beyond these core mechanisms, we also propose a KV sharing variant that reuses KV pairs from the first recursion, specifically designed to further decrease memory footprint. Across model scales ranging from 135M to 1.7B parameters, MoR forms a new Pareto frontier: at equal training FLOPs and smaller model sizes, it significantly lowers validation perplexity and improves few-shot accuracy, while delivering higher throughput compared with vanilla and existing recursive baselines. These gains demonstrate that MoR is an effective path towards large-model quality without incurring large-model cost.

Time-series forecasting is an essential task with wide real-world applications across domains. While recent advances in deep learning have enabled time-series forecasting models with accurate predictions, there remains considerable debate over which architectures and design components, such as series decomposition or normalization, are most effective under varying conditions. Existing benchmarks primarily evaluate models at a high level, offering limited insight into why certain designs work better. To mitigate this gap, we propose TimeRecipe, a unified benchmarking framework that systematically evaluates time-series forecasting methods at the module level. TimeRecipe conducts over 10,000 experiments to assess the effectiveness of individual components across a diverse range of datasets, forecasting horizons, and task settings. Our results reveal that exhaustive exploration of the design space can yield models that outperform existing state-of-the-art methods and uncover meaningful intuitions linking specific design choices to forecasting scenarios. Furthermore, we release a practical toolkit within TimeRecipe that recommends suitable model architectures based on these empirical insights. The benchmark is available at: https://github.com/AdityaLab/TimeRecipe.