Author

Patrick Wong

View on Semantic Scholar

2 papers indexed here

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

Book Open access Sep 2026

Boundary Conditions for Semantic Log Parsing in Systems Observability Pipelines

Parsing decisions in log anomaly detection pipelines directly shape downstream detection effectiveness. Despite growing work on semantic parsers and LLM-assisted parsing, the deployment conditions under which semantic enrichment is preferable to efficient syntactic parsing have not been empirically characterised. Semantic-enriched parsing can improve template distinguishability by using lightweight language-model representations. It also introduces extra representation complexity and embedding inference overhead. Systems operators therefore need evidence about when this additional complexity is useful, negligible, or unstable. This is a systems observability design study, not a parser-proposal paper: the semantic parser is treated as an existing component, and the contribution is evidence about when that component changes downstream detection behaviour enough to justify pilot testing. We empirically characterise these boundary conditions by evaluating a semantic-enriched parser against three syntactic baselines (Drain, IPLoM, and Spell) across five public LogHub datasets spanning HPC, distributed storage, and cloud logs. We use three anomaly detectors with different modelling assumptions: DeepLog, LogAnomaly, and CNN. We report mean and standard deviation over five random seeds, define semantic-enrichment benefit as ΔSE = F1(semantic) – F1(Drain), and also report ΔSE* relative to the best syntactic baseline. We introduce the Template Expansion Ratio (TER), a simple post-parsing diagnostic for how much additional template vocabulary the semantic parser creates relative to Drain. Material gains concentrate on BGL and Spirit, negligible gains occur on HDFS and OpenStack, and Thunderbird shows unstable behaviour under very low anomaly prevalence. The results support cautious, workload-aware use of semantic parsing in observability pipelines rather than treating semantic enrichment as a universally beneficial replacement for syntactic parsers.

Dumo Ngwenya, Patrick Wong, Dhouha Kbaier · Proceedings of the 19th ACM... · 0 citations

#artificial intelligence Preprint Sep 2026

Drift-Aware LLM Routing with Sparse Contexts and Shared Budgets

A multi-model language service must route each request while preserving workload-level budgets for compute, latency, memory, or monetary cost. Two features make this problem materially harder than static model selection. Prompt representations are high dimensional, so only a small subset of embedding directions may predict the incremental value of a model, and both the request mix and the model frontier drift after launches, fine-tunes, quantization changes, and system updates. We formulate nonstationary sparse contextual routing with multiple knapsack constraints and an optional shadow-audit stream that evaluates a small fraction of prompts on several models. We propose Drift-Aware Sparse Routing (DRS). The policy estimates reward and resource use from a rolling audit window, routes using pessimistic reward and optimistic cost estimates, updates resource shadow prices online, and applies a hard meter before commitment. The analysis separates control from statistics. On any event with uniform prediction radii $\{\beta_t\}$, regret against a paced dynamic fluid benchmark is bounded by the sum of the radii, a capacity-buffer term, and an $O(\sqrt{T})$ pacing term. Under a sparse linear model and bounded drift $V_T$, rolling estimation gives \[ \widetilde O\left( T\sqrt{\frac{s}{\rho W}}+WV_T+\sqrt{T} \right), \] where $s$ is sparsity, $\rho$ is the audit rate, and $W$ is the window length. Optimizing $W$ yields the usual stationary $O(\sqrt{sT/\rho})$ rate when $V_T=0$ and a $O(T^{2/3}(s/\rho)^{1/3}V_T^{1/3})$ adaptation term under drift.

Cheung-Hao Lee, Patrick Wong · 0 citations