首页 > 其他分享 >Proj CDeepFuzz Paper Reading: COMET: Coverage-guided Model Generation For Deep Learning Library Test

Proj CDeepFuzz Paper Reading: COMET: Coverage-guided Model Generation For Deep Learning Library Test

时间:2023-09-06 23:24:50浏览次数:45  
标签:layer Lemon MXNet Generation guided Learning sequences COMET

Abstract

背景:已有的方法(Muffin, Lemon, Cradle) can cover at most 34.1% layer inputs, 25.9% layer parameter values, and 15.6% layer sequences.

本文:COMET
Github: https://github.com/maybeLee/COMET
Bug Type: Crash, NaN, inconsistency between the TensorFlow library and the ONNXRuntime library
Task: fuzzing API of DL Libraries
Method:

  1. designs a set of mutation operators and a coverage-based search algorithm to diversify layer inputs, layer parameter values, and layer sequences in DL models
  2. model synthesis

实验:
对象:ONNXRuntime, MXNet, Keras-MXNet, TF2ONNX, ONNX2PyTorch, Keras, TensorFlow, PyTorch
Competitors: Muffin, Lemon, Cradle
效果:

  1. +32 bugs, 21 confirmed, 7 fixed

标签:layer,Lemon,MXNet,Generation,guided,Learning,sequences,COMET
From: https://www.cnblogs.com/xuesu/p/17683605.html

相关文章

  • Proj CDeepFuzz Paper Reading: IvySyn: Automated Vulnerability Discovery in Deep
    Abstract本文:IvySynTask:discovermemoryerrorvulnerabilitiesinDLframeworksBugType:memorysafetyerrors,fatalruntimeerrorsMethod:利用nativeAPIs中静态写明的类型信息在low-levelkernelcode上执行type-awaremutation-basedfuzzingsynthesizeProofof......
  • ChatGLM2 源码解析:`ChatGLMForConditionalGeneration.forward`
    classChatGLMForConditionalGeneration(ChatGLMPreTrainedModel):def__init__(self,config:ChatGLMConfig,empty_init=True,device=None):super().__init__(config)self.max_sequence_length=config.max_lengthself.transformer=C......
  • Proj CDeepFuzz Paper Reading: DeepGauge: multi-granularity testing criteria for
    Abstract本文:DeepGaugeTask:providemulti-granularitytestingcriteriaforDLsystemsMethod:multi-granularitytestingcriteriaforDLsystems:1.k-multisectionNeuronCoverage2.NeuronBoundaryCoverage3.StrongNeuronActivationCoverage4.Top-kN......
  • 机器学习 -> Machine Learning (III)
    来做一些入门题吧.以下大多是kaggle环境.Q1Titanichttps://www.kaggle.com/competitions/titanicimport#ThisPython3environmentcomeswithmanyhelpfulanalyticslibrariesinstalled#Itisdefinedbythekaggle/pythonDockerimage:https://github.com/......
  • Meta-Learning, A Survey
    一、概述 通常在机器学习里,我们需要用大量的数据来训练一个模型;当场景发生改变时,模型就需要重新训练。这显然提升了成本,而人类学习方式与此不同,一个小孩子在学习动物的过程中,学习了很多动物的名称,当某次给他看一些没有见过的动物时,他总能很快的将新动物和别的动物区分开。Meta......
  • aarch64/arm_v8 环境下编译Arcade-Learning-Environment —— ale-py
       condainstallg++=12   cmake../-DCMAKE_BUILD_TYPE=Release-DPYTHON_INCLUDE_DIR=/home/share/xxx/home/software/anaconda3/include-DPYTHON_LIBRARY=/home/share/xxx/home/software/anaconda3/lib/libpython3.11.so-DPython3_EXECUTABLE=/home/share/x......
  • Q-learning and RL implementation
    Aim:Trainamodeltoproperlyplayvintagevideogames...DeepQ-learningAlgo~VeryshortBriefofNotations:{A,pi(Policy),Q(qualityofaction-atastate),R((s,a,s')-Reward,sstatedoingatogotos'andgetaspecificr)} So,ifwew......
  • 迁移学习(CLDA)《CLDA: Contrastive Learning for Semi-Supervised Domain Adaptation》
    Note:[wechat:Y466551|可加勿骚扰,付费咨询]论文信息论文标题:CLDA:ContrastiveLearningforSemi-SupervisedDomainAdaptation论文作者:AnkitSingh论文来源:NeurIPS2021论文地址:download 论文代码:download视屏讲解:click1简介动机:半监督导致来自标记源和目标样本的......
  • 【五期邹昱夫】CCF-A(TIFS'23)SAFELearning: Secure Aggregation in Federated Learning
    "Zhang,Zhuosheng,etal."SAFELearning:SecureAggregationinFederatedLearningwithBackdoorDetectability."IEEETransactionsonInformationForensicsandSecurity(2023)."  本文提出了一种在联邦学习场景下可以保护隐私并防御后门攻击的聚合方法。作者认......
  • Learning Auxiliary Monocular Contexts Helps Monocular 3D Object Detection (2)
    Featurebackbone采用DLA,输入维度为3×H×W的RGB图,得到维度D×h×w的特征图F,然后将特征图送入几个轻量级regressionheads,2Dboudingboxes的中心特征图用下面的模块得到:其中AN是AttentiveNormalization.用公式表示:类似的,2D和3Dboudingboxes的中心之间的offset用公......