Learning Graph-Based Heuristics for Pointer Analysis without Handcrafting Application-Specific Featu

on YouTube ↗15m

" summary

Graphick automatically generates graph-based context-sensitivity and heap-abstraction heuristics by learning feature sequences over program graphs (object-allocation and field points-to graphs) and validating with the Doop Java pointer analyzer. Features are topological patterns of nodes expressed as incoming/outgoing edge intervals; learning enumerates splits and predecessor/successor extensions using a score function and a gamma hyperparameter to avoid overfitting. Graphick matches or outperforms Scaler and Zipper on many benchmarks, beats prior ML-based Data on precision, and rescues bloat/jedit that other heuristics time out on within 3 hours using Graphick and Doop.

AI-written summary · about

" chapters (16)

1. 0:00Intro
2. 0:20Importance of Pointer Analysis Pointer analysis is a key component of various software engineering tools
3. 0:54Analysis Heuristics in Pointer Analysis . Cost-effective pointer analyzer contains various analysis heuristics Precise & scalable analysis result
4. 1:20Necessity of Context Sensitivity Heuristics
5. 2:47Current Trend on Designing Heuristics
6. 4:08Problem
7. 4:37Our Goal
8. 5:47Our Technique: Graphick
9. 6:28How a Learned Heuristic Works
10. 8:54How a Feature Work
11. 10:02Learning a Set of Features Our algorithm transforms all the labeled node into a set of features
12. 10:42Learning a Feature
13. 12:17Our Framework: Graphick
14. 12:32Settings
15. 13:25Comparison to a Previous Data-driven Context Sensitivity Heuristic
16. 14:11Summary . We made Graphick to automatically generate graph-based analysis heuristics Two key ideas are our feature description language and learning algorithm Graphs Analyzer Graphick

" related talks

• 15mGraphRAG methods to create optimized LLM context windows for RetrievalJonathan Larson · AI Engineer World's Fair 2025
• 01h 41mGraph Intelligence: Enhance Reasoning and Retrieval Using Graph AnalyticsAndreas, Alison · AI Engineer World's Fair 2025
• 14mA Practitioner's Guide to GraphsTim Ainge · AI Engineer World's Fair 2026
• 30mScalable Graph-Parallel Inference Algorithms and SystemsJoseph E. Gonzalez · Parallel and Distributed Algorithms for Inference and Optimization (Simons Institute, Oct. 21–24, 2013)
• 12mTopology Over Predictions: Using Graph Context to Drive Real-Time Lending DecisionsMatthew Watts · NODES 2025
• 15mGauss: Program Synthesis by Reasoning Over GraphsIon Stoica, Caroline Lemieux, Koushik Sen, Rohan Bavishi · OOPSLA 2021
• 06mCloud-Native GraphRAG: AI’s Logical EdgeStephen Chin · Cloud Native + Kubernetes AI Day North America 2025
• 29mTed Willke, Senior Principal Engineer & GM, Datacenter Group, IntelTed Willke · MLconf SF 2014