Tianyi Li

Graph Neural Architecture Search (GNAS), a subdomain of Automated Machine Learning (AutoML), aims to automate the design of graph neural network (GNN) architectures. However, GNAS is an inherently data-intensive process, as it calls for the training and evaluation of numerous candidate GNN architectures, leading to massive data access redundancy. Existing multi-GPU GNAS frameworks typically use data parallelism, which incurs high synchronization costs, or architecture parallelism, which struggles to scale to large datasets. We propose FastGNAS, a fast and scalable multi-GPU framework designed to tackle the data management challenges underlying GNAS.First, FastGNAS introduces a hybrid parallel framework that combines data and architecture parallelism, enabled by a novel ring-based model migration. Specifically, it offers an efficient data flow that exploits scalable data parallelism and isolated architecture parallelism to enable synchronization-free processing. Next, to address data redundancy, FastGNAS offers a specialized caching layer for intermediate data products, implemented through advanced batch management strategies including incremental storage and a probabilistic reuse policy. Finally, FastGNAS employs fine-grained load balancing and scheduling via exploratory task generation and predictive workload estimation, enabling purposeful resource allocation across candidate architectures. Extensive experiments show that FastGNAS can accelerate state-of-the-art baselines by an average of 3.14× (up to 6.18×) on diverse benchmarks, while achieving competitive accuracy.

Multi-modal analytical processing has the potential to transform applications in e-commerce, healthcare, entertainment, and beyond. However, real-world adoption remains elusive due to the limited ability of traditional relational query operators to capture query semantics. The emergence of foundation models, particularly the large language models (LLMs), opens up new opportunities to develop flexible, semantic-aware data analytics systems that transcend the relational paradigm. We present Nirvana, a multi-modal data analytics framework that incorporates programmable semantic operators while leveraging both logical and physical query optimization strategies, tailored for LLM-driven semantic query processing. Nirvana addresses two key challenges. First, it features an agentic logical optimizer that uses natural language-specified transformation rules and random-walk-based search to explore vast spaces of semantically equivalent query plans --- far beyond the capabilities of conventional optimizers. Second, it introduces a cost-aware physical optimizer that selects the most effective LLM backend for each operator using a novel improvement-score metric. To further enhance efficiency, Nirvana incorporates computation reuse and evaluation pushdown techniques guided by model capability hypotheses. Experimental evaluations on three real-world benchmarks demonstrate that Nirvana is able to reduce end-to-end runtime by 10%–85% and reduces system processing costs by 76% on average, outperforming state-of-the-art systems at both efficiency and scalability.

Constrained Reinforcement Learning (CRL) aims to optimize decision-making policies under constraint conditions, making it highly applicable to safety-critical domains such as autonomous driving, robotics, and power grid management. However, existing robust CRL approaches predominantly focus on single-step perturbations and temporally independent adversarial models, lacking explicit modeling of robustness against temporally coupled perturbations. To tackle these challenges, we propose TCRL, a novel temporal-coupled adversarial training framework for robust constrained reinforcement learning (TCRL) in worst-case scenarios. First, TCRL introduces a worst-case-perceived cost constraint function that estimates safety costs under temporally coupled perturbations without the need to explicitly model adversarial attackers. Second, TCRL establishes a dual-constraint defense mechanism on the reward to counter temporally coupled adversaries while maintaining reward unpredictability. Experimental results demonstrate that TCRL consistently outperforms existing methods in terms of robustness against temporally coupled perturbation attacks across a variety of CRL tasks.