Research

Pipeline parallelism can improve prefill throughput by processing multiple request chunks concurrently across different stages of the model. However, keeping the pipeline fully utilized requires efficient scheduling and request preparation. In systems where stages retain and evict cache state independently, a local cache hit does not guarantee that the same prefix can be reused across the pipeline. Here, coordination overhead can impede request admission cadence and thus reduce overall throughput. In this paper, we present WavePP, a prefill runtime built on top of TensorRT-LLM that addresses these challenges by overlapping request admission with pipeline execution. WavePP asynchronously finds a prefix that can be reused across all stages, protects the cached state, and reserves space for the remaining input while earlier requests continue to execute. It subsequently plans the chunk sizes of each request dynamically to maximize pipeline fill. Each stage then completes the local preparation before executing the request. In the same system and pipeline topology, WavePP improves TensorRT-LLM's prefill throughput in 37 of 40 tested settings on GLM 5.2 and MiniMax M2.7. At concurrency 128 with high cache reuse, these changes increase throughput by factors of 2.91 and 2.02, respectively. Across 28 Kimi K3 settings, WavePP also has the highest measured throughput in all 18 settings at concurrency eight or higher, compared with tensor/expert-parallel and pipeline-parallel baselines from TRT-LLM, SGLang, and vLLM.

Speculative decoding accelerates language model inference by using a fast drafter to propose candidate tokens that are then verified by a larger target model. Existing theory largely studies the stochastic, distribution-preserving setting, where the goal is to exactly sample from the target distribution. In contrast, many practical systems use greedy decoding, relaxed acceptance rules, or tree-based candidate sets, where success is governed by local ranking and threshold events rather than exact distributional equality. We develop a theory for these regimes. We identify that many common acceptance criteria have rejection regions that can be characterized as lower level sets of the target distribution. For these, we characterize the exact KL divergence required for rejection yielding exact certificates and sharp margin-based bounds for strict greedy decoding, additive and multiplicative relaxed acceptance, top-(m) relaxed criteria, and entropy-thresholded acceptance. We then extend the framework to greedy tree decoding, deriving exact and margin-only certificates for when the target greedy token remains covered by the drafter's top-(m) candidates. Finally, we evaluate the resulting certificates on Qwen3 models, showing that relaxed and tree-based criteria substantially enlarge the region of certified acceptance, especially on decoding steps with low target model distribution margin. These results complement existing distribution-preserving analyses of speculative decoding by characterizing the deterministic local acceptance events common in practical inference systems.

We propose integrating topological data analysis (TDA) with deep learning for satellite image classification. TDA extracts geometric features through persistence homology while addressing CNNs' tendency toward texture-based local features. The method achieves 99.33% accuracy on the EuroSAT dataset, exceeding prior results from larger models like ResNet50 and XL Vision Transformers. It also demonstrates 1.82% improvement over ResNet18 baseline on RESISC45. This represents the first application of TDA features in satellite scene classification combined with deep learning.

We tackle air pollution monitoring by predicting air-quality index values in 1 km² neighborhoods using the AirDelhi dataset. Employing spatio-temporal graph neural networks, we achieve a 79% reduction in error compared to previous approaches, with performance that generalizes to previously unseen locations. The work also identifies new patterns in AQI behavior, including repetitive short-term fluctuations and evolving spatial relationships in pollution distribution.

We present a framework for enhancing multi-agent cooperation through normative modules. These modules enable agents powered by large language models to recognize and adapt to the normative infrastructure of a given environment. The work addresses equilibrium selection in cooperation by drawing on game theory concepts like correlated equilibrium. Agents learn which institutions are considered authoritative through peer interactions, allowing them to coordinate sanctioning behaviors and achieve more stable cooperative outcomes. The evaluation demonstrates the system's capacity to disregard non-authoritative institutions and identify legitimate ones among alternatives.

We propose a computer vision system for monitoring dietary consumption using stationary cameras positioned to observe food on utensils. Rather than requiring users to photograph their completed meals from above, our method tracks food items on spoon surfaces in real-time. The approach proves particularly effective for mixed foods like soups and stews where ingredients submerged in a stew would otherwise remain invisible. We frame this as a non-invasive, user-friendly, and highly accurate dietary intake monitoring tool that eliminates perspective limitations of conventional meal photography approaches.

This paper presents a polynomial time recognition algorithm for (p,q)-split graphs, developed during a research term in Graph Theory under the supervision of Professor Sophie Spirkl at the University of Waterloo (January 2023 to April 2023).