Research Topic 03 · Diagnostics Online

Software Engineering for AI

Testing, diagnosing, and localizing faults in AI systems — from deep learning programs and libraries to machine translation and coreference resolution.

3
Papers
6
Venues
ACM SIGSOFT
Outstanding Dissertation
Award
Scroll to run diagnostics
Featured Work · Fault Diagnosis
ICSE '22

DeepFD: Automated Fault Diagnosis and Localization for Deep Learning Programs

Jialun Cao, Meiziniu Li, Xiao Chen, Ming Wen, Yongqiang Tian, Bo Wu, Shing-chi Cheung

Deep learning programs fail in silent, non-crashing ways — a wrong loss function, a learning rate that's too large, or a mismatched activation. DeepFD frames diagnosis as a three-step learning problem — diagnostic feature extraction, fault diagnosis, and fault localization — correctly diagnosing 52% of faulty programs (vs. 27% for prior state-of-the-art) and localizing 42% (vs. 23%).

💡 DeepFD correctly diagnoses 52% of faulty DL programs (vs. 27% SOTA) and localizes 42% of faults (vs. 23%) using a learning-based three-step pipeline.
DeepFD Diagnostic Console idle
Featured Work · Metamorphic Testing
TOSEM '22

SemMT: Semantic-Based Testing for Machine Translation Systems

Jialun Cao, Meiziniu Li, Yeting Li, Ming Wen, Shing-chi Cheung

Without reference translations, how do you know a translator is wrong? SemMT round-trips a sentence through a translation system and measures the semantic distance between input and output — capturing logical and numeric constraints via regex and DFA — when meaning drifts, a bug is exposed. SemMT improves accuracy by 21% and F-score by 23% over prior work.

💡 Round-trip semantic distance — capturing logical and numeric constraints via regex and DFA — exposes translation bugs with 21% better accuracy and 23% better F-score.
Source (EN) The bank is by the river.
▁▁▁ round-trip translation ▁▁▁
G Translator A
EN → ZH → EN
ZH河岸边有一家银行... → 河边有河岸
EN'The riverbank is by the river.
semantic similarity
✓ Meaning preserved
D Translator B
EN → ZH → EN
ZH银行在河边 → 银行 (financial)
EN'The financial institution is by the river.
semantic similarity
⚠ Semantic drift — bug found
Featured Work · Unlabeled Testing
FSE '23

Testing Coreference Resolution Systems without Labeled Test Sets

Jialun Cao, Yaojie Lu, Ming Wen, Shing-Chi Cheung

Coreference systems link pronouns to the entities they refer to. Crest builds metamorphic test pairs from constituency and dependency relations that must preserve the same coreference — with no labeled test set required — and flags a bug whenever the predicted links become inconsistent across the pair. On 1,000 CoNLL-2012 sentences, Crest reaches 100% precision, versus 63–75% for prior baselines.

💡 Metamorphic test pairs built from constituency and dependency relations achieve 100% precision on coreference bugs with zero labeled test data.
Original passage bugs found: 0
John hired Mary because he believed she was the most talented engineer on the team.
Entity A links Entity B links Inconsistency (bug)