08 2023 档案
摘要:## Abstract Github: https://github.com/marsggbo/automl_a_survey_of_state_of_the_art 本文: 1. intro AutoML methods: data preparation, feature engineering
阅读全文
摘要:## Abstract 背景: 1. the de facto standard to assess the quality of DNNs in the industry is to check their performance (accuracy) on a collected set of
阅读全文
摘要:## Abstract ## 1. Intro ## 2. Background ### 2.1 Program Understanding and Generation Tasks ### 2.2 NL-PL Pre-Trained Models 这类的函数崩溃 本文:GRIST Github: https://github.com/Jacob-yen/GRIST Task: generate a small
阅读全文
摘要:## Abstract 本文: Task: formalize the space of adversaries against DNNs and then introduce an adversarial testing 实验: 方法:defining a hardness measure 效果:
阅读全文
摘要:## Abstract 背景: the robustness requires the model to produce consistent decisions given minorly perturbed code inputs 本文:CARROT Github: https://github
阅读全文
摘要:## Abstract
阅读全文
摘要:## Abstract 本文: DeepPERF Github: https://dlperf.github.io/ 对象:TensorFlow, Keras Task: 1. characterize symptoms, root causes, and introducing and expos
阅读全文
Proj CDeepFuzz Paper Reading: A Brief Introduction to Automatic Differentiation for Machine Learning
摘要:## Abstract 本文:描述AD, its motivations, implementation approaches, dataflow programming, example programs with Tensorflow and PyTorch ## 1. Intro ## 2.
阅读全文
摘要:## Abstract 背景: 1. AD涉及computational fluid dynamics, atmospheric sciences, engineering design optimization 2. AD与DL长时间并不相互交流,直到dynamic computational g
阅读全文
摘要:## Abstract 背景:在architecture level detecting bugs获利更高 本文:DEBAR Github: https://github.com/ForeverZyh/DEBAR Task: static analysis of neural architectur
阅读全文
摘要:## Abstract 本文: Pythia Task: use static analysis to track the shapes of tensors across Python library class and warns of several possible mismatches M
阅读全文
摘要:## Abstract 本文:review on autoML emphasis on unsupervised anomaly detection ## 1. Intro ## 2. AutoML ### 2.1 Challenges ### 2.2 Model generation ## 3.
阅读全文
摘要:## Abstract 背景:由于收集到的数据,模型会学习到对性别、职业、国籍、种族的偏见,需要发现这些fairness bugs 本文:BiasFinder Task: Discover fairness bugs in the Sentiment Analysis System via Meta
阅读全文
摘要:## Abstract 背景:关于test RNN-based stateful system的研究较少 本文:DeepStellar Method: 1. model RNN as an abstract state transition system 2. design 2 trace simi
阅读全文
摘要:## Abstract 本文:VerifyML Task: testing ML based applications(ML based image classifier) by metamorphic testing Github: https://github.com/verml/VerifyM
阅读全文
摘要:## Abstract 本文: 1. 分析DL Library中的错误,给出了2个研究问题的答案 1. RQ1: TensorFlow 中的错误有哪些症状和原因? 2. RQ2: TensorFlow 内部的错误在哪⾥? 2. 总结了5个发现 1. 与症状相⽐,根本原因更具决定性,因为多个根本原因主
阅读全文
摘要:## Abstract 本文:RTIs(Referentially Transparent Inputs), Purity Task: Testing Machine Translation Model Method: referentially transparent input(在不同的上下文中
阅读全文
摘要:## Abstract Background: 1. discriminatory inputs, e.g., from societal bias, produce error, need to conduct fairness testing(generating discriminatory
阅读全文
Proj CDeepFuzz Paper Reading: Detecting TensorFlow program bugs in real-world industrial environment
摘要:## Abstract 本文: ShapeTracer Task: 1. Study on 12289 failed TensorFlow jobs 2. detecting TensorFlow shape-related errors with a constraint-based approa
阅读全文
摘要:## Abstract 本文:Metamorphic Testing for machine learning classifiers Method: cross-validation, metamorphic testing 1. MR-0: Consistence with affine tra
阅读全文
摘要:## Abstract 背景:目前的研究集中在例如交换几个像素这种digital perturbation,这些是很难发生在现实世界的;生成both digital and physical adversarial perturbation很重要 本文:DeepBillboard Task: Tes
阅读全文
摘要:## Abstract 本文:Eagle Task: use equivalent graphs to differential test DL libraries Method: 使用不同的APIs, data types or optimizations来获取equivalent graphs,
阅读全文
摘要:## Abstract 背景:Neural networks can be regarded as a new programming paradigm, Tensorflow 和PyTorch相当于Compiler,我们已知如果编译器缺乏Specification 本文:ExAIS Task: p
阅读全文
摘要:## Abstract 本文: KONURE,以及regenerator Task: use active learning to infer models of applications that retrieve data from relational databases 方法: 1. a d
阅读全文
摘要:## Abstract 背景: 1. Q: active learning inference based framework能够利用modularity来处理large applications中的develeopment correctness, performance and cost 2.
阅读全文
摘要:## Abstract 本文: RULF Github: https://github.com/Artisan-Lab/RULF Task: Library harness generation for Rust Library via API dependency graph traversal
阅读全文
摘要:## Abstract 背景: 1. 软件供应链攻击的目标是集成到客户端应用程序中的组件。 2. 此类攻击通常针对广泛使用的组件,通过不影响客户端观察到的组件行为(例如文件系统或网络访问)进行攻击。 本文:HARP Task: infer and regenerate the client-obse
阅读全文
摘要:## Abstract 背景: 1. unsafe能够绕开rust type system 2. rust libraries中常有许多unsafe keyword 本文:SyRust Task: fuzz Rust library APIs Challenge: synthesize well-t
阅读全文
摘要:## Abstract 背景:现有的深度学习测试在很⼤程度上依赖于⼿动标记的数据,因此通常⽆法暴露罕⻅输⼊的错误⾏为。 本文:DeepXplore Task: a white-box framework to test DL Models 方法: 1. neuron coverage 2. diff
阅读全文
摘要:## Abstract 背景: 1. 深度学习库需要满足编程语言的语法、语义需求,也需要满足valid computational graphs的tensor/operator constraints 2. LLM倾向于生成普通正常的程序,但是模糊测试则需要异常模式 3. 历史错误可能更有价值 本文
阅读全文
摘要:## Abstract 本文: DLFuzz 方法: Adversarial inputs + Fuzzing, Differential Testing between before/after adding perturb 实验: 数据集:MNIST LeNet, ImageNet + VGG/
阅读全文
摘要:## Abstract 本文:DLFuzz Task: Fuzz Machine Learning Models using differential testing to detect inconsistency before/after perturbation with minutely mu
阅读全文
摘要:## Abstract 背景: 1. 对微小扰动十分脆弱主要来自于overfitting、limited datasets 2. 由于Specification往往不完备,因此automatic program repair通常要从程序行为中学习 本文:SENSEI, SENSEI-SA Githu
阅读全文
摘要:## Abstract 背景:已有的方法仅仅利用了已经存在的models,只在model inference阶段检测bugs 本文: Muffin 方法:specifically-designed model fuzzing approach + 定制一组指标来衡量不同DL库之间的不一致differ
阅读全文
摘要:## Abstract 本文:CocoFuzzing Task: test ML Models, test code processing models 方法:10 mutators to automatically generate validly and semantically preserv
阅读全文
摘要:## Abstract 背景:Python API使得DL Libraries的参数类型难以确定 本文:FreeFuzz Github: https://github.com/ise-uiuc/FreeFuzz 方法:从1. 来自库文档的代码片段 2. 测试 3. DL models in the
阅读全文
摘要:## Abstract 本文: RapidFuzz Task: use GAN to fuzz program with highly-structured inputs Method: use GAN to capture the data structure, then locate the h
阅读全文
摘要:## Abstract 本文:graph-based fuzz testing Task: Fuzzing Machine Learning Libraries using operator graph Github: https://github.com/gbftdlie/G (Q?https:/
阅读全文
摘要:## Abstract 背景: 1. Q: 现有的方法没有考虑到small perturbations的影响,或者这些perturbation只能限定在某个特定的模型上使用,在其他模型上则本身就不符合example要求(可能是指标签就会改变?) 2. 现有的方法多使用浅层的feature const
阅读全文
摘要:## Abstract 本文: DeepHunter Task: Fuzzing Deep Learning Models Github: https://github.com/Shimmer93/Deephunter-backup Method: 1. Metamorphic mutation t
阅读全文
摘要:## Abstract 背景: 1. 目前已有工作test underlying operator-level DL Libraries,但测试compier的工作比较少 2. compiler: compile high-level tensor computation graphs into h
阅读全文
摘要:## Abstract 背景:现有的constraint-extraction tech 对extracting DL-specific input constraints效率太低 本文:DocTer Github: https://github.com/lin-tan/DocTer Task: e
阅读全文
Proj CDeepFuzz Paper Reading: Fuzzing Deep-Learning Libraries via Automated Relational API Inference
摘要:## Abstract 背景:使用API间sharing similar input parameters and inputs的关系能更有效测试DL Libraries 本文: DeepREL Github: https://github.com/ise-uiuc/DeepREL Task: in
阅读全文
摘要:## Abstract 背景: 1. In property-based testing(PBT), one asserts properties that a function should satisfy and the system automatically generates tests
阅读全文
摘要:## Abstract 背景:马尔可夫决策过程(Markov decision process, MDP)是串联决策问题(sequential decision making)的一种数学化建模;机器学习已经为MDP提供了很多解法,但这些解法没有被严格测试过,或者不真正可靠(Q?) 本文:MDPFuz
阅读全文
摘要:## Abstract 背景:Deep-learning compiler例如TVM和TensorRT被使用来optimize DNN模型以达到更好的性能要求;在Deep Learning Compiler中的bug可能导致语义改变 本文:NNSmith Task: fuzz deep-learni
阅读全文
摘要:## Abstract 本文:TitanFuzz Task: use LLM to fuzz DL librarires Method: use both generative and infilling LLMs(e.g., Codex, InCoder) 实验: 对象:Tensorflow, P
阅读全文
摘要:## Abstract 背景: 挑战: 1. need valid input domain of each API function 2. hard to trigger new behavior 本文:SkipFuzz 任务: fuzz machine learning libraries us
阅读全文
摘要:## Abstract 本文:ConFL Task: Fuzz Deep Learning Libraries using Constraint-guided fuzzer Method: Constraint-guided fuzzer + grouping 实验: 对象:Tensorflow(M
阅读全文
摘要:## Abstract 本文:DeFuzz Task: guided directed fuzzing Method: pre-trained BiLSTM to identify the potentially vulnerable functions, directed fuzzing 实验:
阅读全文
摘要:## Abstract 背景:在最近一项bug study中,high-level IR的优化导致44.92%的bug(Q?) 本文: HirGen Github: https://github.com/haoyang9804/HirGen Task: fuzzing the optimizatio
阅读全文
摘要:## Abstract 本文:DRFuzz Task: find the regression faults between versions of a DL system Method: 1. a diversity-oriented test criterion to explore as ma
阅读全文
摘要:## Abstract 背景:每个深度学习库 API 都可以抽象为处理张量/向量的函数(each DL library API can be abstracted into a function processing tensors/vectors),可用来差分测试 本文:∇Fuzz Task: A
阅读全文
摘要:## Abstract 本文: DPFuzz Github: https://github.com/Tizpaz/DPFuzz Task: Differential performance analysis with input groups, provide an explanation for
阅读全文