arXiv subjects

Raina Wu

Publications and source records attributed to Raina Wu.

4 recordsLinked to original sources

arXiv2025

Tailored Primitive Initialization is the Secret Key to Reinforcement Learning

Reinforcement learning (RL) has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models (LLMs). While RL has demonstrated substantial performance gains, it still faces key challenges, including low sampling efficiency and a strong dependence on model initialization: some models achieve rapid improvements with minimal RL steps, while others require significant training data to make progress. In this work, we investigate these challenges through the lens of reasoning token coverage and argue that initializing LLMs with diverse, high-quality reasoning primitives is essential for achieving stable and sample-efficient RL training. We propose Tailor, a finetuning pipeline that automatically discovers and curates novel reasoning primitives, thereby expanding the coverage of reasoning-state distributions before RL. Extensive experiments on mathematical and logical reasoning benchmarks demonstrate that Tailor generates more diverse and higher-quality warm-start data, resulting in higher downstream RL performance.

cs.LG↗

arXiv2025

Influence Prediction in Collaboration Networks: An Empirical Study on arXiv

This paper provides an empirical study of the Social Sphere Model for influence prediction, previously introduced by the authors, combining link prediction with top-k centrality-based selection. We apply the model to the temporal arXiv General Relativity and Quantum Cosmology collaboration network, evaluating its performance under varying edge sampling rates and prediction horizons to reflect different levels of initial data completeness and network evolution. Accuracy is assessed using mean squared error in both link prediction and influence maximization tasks. The results show that the model effectively identifies latent influencers, i.e., nodes that are not initially central but later influential, and performs best with denser initial graphs. Among the similarity measures tested, the newly introduced RA-2 metric consistently yields the lowest prediction errors. These findings support the practical applicability of the model to predict real-world influence in evolving networks.

cs.SI↗

arXiv2025

The Social Sphere Model: Heuristic Influence Prediction in Evolving Networks

How would admissions look like in a university program for influencers? In the realm of social network analysis, influence maximization and link prediction stand out as pivotal challenges. Influence maximization focuses on identifying a set of key nodes to maximize information dissemination, while link prediction aims to foresee potential connections within the network. These strategies, primarily deep learning link prediction methods and greedy algorithms, have been previously used in tandem to identify future influencers. However, given the complexity of these tasks, especially in large-scale networks, we propose an algorithm, The Social Sphere Model, which uniquely utilizes expected value in its future graph prediction and combines specifically path-based link prediction metrics and heuristic influence maximization strategies to effectively identify future vital nodes in weighted networks. Our approach is tested on two distinct contagion models, offering a promising solution with lower computational demands. This advancement not only enhances our understanding of network dynamics but also opens new avenues for efficient network management and influence strategy development.

cs.SI↗

arXiv2025

AI Agents in Engineering Design: A Multi-Agent Framework for Aesthetic and Aerodynamic Car Design

We introduce the concept of "Design Agents" for engineering applications, particularly focusing on the automotive design process, while emphasizing that our approach can be readily extended to other engineering and design domains. Our framework integrates AI-driven design agents into the traditional engineering workflow, demonstrating how these specialized computational agents interact seamlessly with engineers and designers to augment creativity, enhance efficiency, and significantly accelerate the overall design cycle. By automating and streamlining tasks traditionally performed manually, such as conceptual sketching, styling enhancements, 3D shape retrieval and generative modeling, computational fluid dynamics (CFD) meshing, and aerodynamic simulations, our approach reduces certain aspects of the conventional workflow from weeks and days down to minutes. These agents leverage state-of-the-art vision-language models (VLMs), large language models (LLMs), and geometric deep learning techniques, providing rapid iteration and comprehensive design exploration capabilities. We ground our methodology in industry-standard benchmarks, encompassing a wide variety of conventional automotive designs, and utilize high-fidelity aerodynamic simulations to ensure practical and applicable outcomes. Furthermore, we present design agents that can swiftly and accurately predict simulation outcomes, empowering engineers and designers to engage in more informed design optimization and exploration. This research underscores the transformative potential of integrating advanced generative AI techniques into complex engineering tasks, paving the way for broader adoption and innovation across multiple engineering disciplines.

cs.AI↗