SlideShare a Scribd company logo
1 of 52
1
A Survey on Security and Privacy of
Machine Learning
Dang Duy Thang
September 15th 2019
2
Big thanks to VJAI community and
BizReach company for supporting this
talk!
3
Outline
1. Introduction
2. Adversarial examples on CV
3. Adversarial examples on NLP
4. Adversarial examples on Speech
5. Conclusion
4
Introduction
What can AI learn from security?
• History has shown attacker always follows footsteps of
new technology development.
• As AI controls more and more systems, attacker will
have higher & higher chances.
5
Failures of ML
Microsoft’s AI chatbot
The bot was Artificial Intelligence
Chatterbot that created by Microsoft, and
named "Tay" after the acronym "thinking
about you".
There are similar platforms as Xiaoice in
China and Rinna in Japan.
• Tay was designed to mimic the language patterns of a 19-year-old
American girl, and to learn from interacting with human users of
Twitter.
• The system collected vast amounts of intimate details on
individuals, the program raises privacy questions.
6
Failures of ML
Microsoft’s AI chatbot
Microsoft had to shut down “TAY” only
16 hours after its launch on March 23,
2016 because it started tweeting Racist
messages such as:
“Tay accused George W. Bush of causing 9/11, praise
Hitler, refered to President Barack Obama as a “monkey”
Tay's behavior on a "coordinated attack by a subset of
people" that "exploited a vulnerability in Tay.
7
What is the adversarial examples?
In the paper “Explaining and harnessing Adversarial
Examples”, Goodfellow et al, International Conference on
Learning Representations 2015 :
“Adversarial examples is that ML models misclassify examples
that are only slightly different from correctly classified examples
draw from the data distribution”.
8
Adversarial Examples
Research fields are effected by Adversarial Examples
1. Computer vision
2. Natural language processing
3. Speech processing
9
Adversarial Examples is a big problem?
https://nicholas.carlini.com/writing/2019
/all-adversarial-example-papers.html
NeurIPS - Workshop on Security in Machine
Learning 2019
ICML - Workshop on the Security and Privacy
of Machine Learning 2019
ICLR - Safe Machine Learning Specification,
Robustness and Assurance Workshop 2019
CVPR - The Bright and Dark Sides of Computer
Vision: Challenges and Opportunities for
Privacy and Security Workshop 2019
https://github.com/IBM/adversar
ial-robustness-toolbox https://github.com/tensorflow/cleverhans
10
Adversarial Examples
There are two types of adversarial examples attacks:
1. White-box attacks
2. Black-box attacks
Or we can categorize AE as output type:
1. Untargeted attack
2. Targeted attack
11
Image Classification
http://cs231n.github.io
Each number is an integer that ranges
from 0 (black) to 255 (white)
12
Interpreting a linear classifier
Score function that maps the raw image pixels
to class scores
Higher score means better, or ('more likely')
Loss function: measures quality of outcomes
the loss will be high if we’re doing a poor classification, and it
will be low if we’re doing well.
http://cs231n.github.io
13
72 136 56 103
61 52 134 42
142 210 134 99
150 46 81 76
72 136 56 201
61 52 134 172
142 210 134 81
48 35 110 200
How AI system recognizes a RGB Image?
Input RGB Image
“vending machine”
Red channel
Green channel
Blue channel
72 136 56 80
61 52 134 97
142 210 134 178
36 79 63 85
What a computer sees
Three matrices correspond to three channels
3つの行列が3つのチャンネルに対応
14
72 136 56 103
61 52 134 42
142 210 134 99
150 46 81 76
72 136 56 201
61 52 134 172
142 210 134 81
48 35 110 200
How AI system recognizes a RGB Image?
72 136 56 80
61 52 134 97
142 210 134 178
36 79 63 85 Flatten
72
136
56
80
…
…
201
172
81
…
…
103
42
99
…
3 matrices
Vector
vectorization
15
How AI system recognizes a RGB Image?
72
136
56
…
…
…
…
…
201
…
…
…
103
…
…
Vector Input Layer
Hidden Layer
Output Layer
thatch
keyboard
…
16
AI Model
Let’s consider the Google Inception V3 that was
trained on ImageNet dataset with 1,000 classes. Input
image is a color image with 299x299 size.
Input RGB Image 299 x 299
72 136 56 103
61 52 134 42
142 210 134 99
150 46 81 76
72 136 56 201
61 52 134 172
142 210 134 81
48 35 110 200
72 136 56 80
61 52 134 97
142 210 134 178
36 79 63 85
299 x 299 x3
1x268,203
scaling
Vectorization
72
136
56
80
…
…
201
172
81
…
…
103
42
99
…
17
Normal process
In a normal process to train a neural network:
ニューラルネットワークを訓練する通常のプロセ
スでは
Fix, no change
固定、変更なし
Fine-tune W respect to Loss(input, label)
Try to find the best W that makes Loss function is as small as possible
損失に対するWの微調整(入力、ラベル)
損失関数ができるだけ小さくなるような最良のWを見つけようとする
label
http://cs231n.github.io
18
Crafting Adversarial Examples
In a process to create adversarial examples:
敵対的な例を作成する過程で
Fine-tune input respect to Loss(input, label)
Keep Loss(input, label) is large
Keep Loss(adversarial, new label) is small
label
Fix, no change
固定、変更なし
損失に対する入力の微調整(入力、ラベル)
(入力、ラベル)の損失は大きく、
(敵対的、新しいラベル)の損失は小さくする
http://cs231n.github.io
19
Adversarial Attack
Input vector
AI System
46
140
37
92
…
…
118
37
41
…
…
90
29
30
…
Adversarial Vector
72 136 56 90
61 52 134 29
142 210 134 30
150 46 81 76
72 136 56 118
61 52 134 37
142 210 134 41
48 35 110 200
46 140 37 92
61 52 134 97
142 210 134 178
36 79 63 85
299 x 299 x3
72
136
56
80
…
…
201
172
81
…
…
103
42
99
…
Adversarial Image
reform
20
Back to FGSM (Fast Gradient Sign Method)
FGSM (Fast Gradient Sign Method)*:
(*) Ian Goodfellow et al, Explaining and harnessing adversarial examples, ICLR 2015.
72
136
56
80
…
…
201
172
81
…
…
103
42
99
…
46
140
37
92
…
…
118
37
41
…
…
90
29
30
…
= + adversarial perturbation
21
Adversarial Examples is actually dangerous?
Paper: Fooling automated surveillance cameras:
adversarial patches to attack person detection, CVPR
2018
22
Adversarial Examples is actually dangerous?
Autonomous car are built on ML/AI for auto detect traffic
sign?
What happen if that car’s AI system does misrecognition
“STOP” traffic sign instead of “turn right” sign when it is
running in a high way?
Taesoo Kim, AI & Security, Microsoft Research Faculty Summit 2017
23
However, white-box attack is not really realistic
because you’re unlikely to get access to the
gradients of the loss function on any particular
system.
-> The paper from Papernot et al, “Practical Black-
Box Attacks against Machine Learning”, ASIA CCS
2017.
Adversarial Samples –
Black-box attacks
24
 In a more realistic context, you would want to
attack a system having only access to its
outputs.
 In the paper “Practical Black-Box Attacks
against Machine Learning” proposed a
substitute Deep Neural Networks to solve the
same classification tasks as the target model.
Adversarial Samples –
Black-box attacks
25
There are two main strategies in black-box attacks:
1. Create a synthetic dataset.
2. Build and Train a local substitute Deep Neural
Networks.
Adversarial Samples –
Black-box attacks
26
How to create a synthetic dataset (input, output)?
1. Inputs: are synthetic and generated by the
adversary.
2. Outputs: are labels assigned by the target DNN
and observed by the adversary.
Adversarial Samples –
Black-box attacks
27
How to build a local substitute Deep Neural Networks?
The attacker queries the oracle (target system) with
synthetic inputs selected by a Jacobian-based
heuristic to build a model F approximating the
oracle model O’s decision boundaries.
Adversarial Samples –
Black-box attacks
28
What are the Oracle DNNs?:
1. Amazon oracle: https://aws.amazon.com/machine-
learning
2. Google oracle: https://cloud.google.com/prediction
3. MetaMind oracle: https://einstein.ai/
They provide some functionalities: dataset upload,
automated model training and model prediction
querying.
This method made MetaMind misclassify at 84.24%,
Amazon and Google at 96.19% and 88.94%.
Adversarial Samples –
Black-box attacks
29
In an another real scenario, the attacker would not be
allowed to provide its own image files, the neural
network would take camera pictures as input. That’s
the problem, the authors of this paper “Adversarial
examples in the physical world”(ICLR, 2017) are
trying to solve.
Adversarial Samples –
Black-box attacks
30
“We used images taken from a cell-phone camera as a input
to an image classification neural network”
Adversarial Samples –
Black-box attacks
31
Adversarial Examples in Natural
Language Processing
32
Adversarial Examples on NLP
 Paper: Deep Text Classification Can be Fooled, Bin Liang et al,
International Joint Conference on Artificial Intelligence, 2018.
 The foundation of attack lies in identifying the text items that
possess significant contribution to the classification by leveraging
the cost gradient.
 They determine the most frequent phrases, called Hot Training
Phrases (HTPs).
33
Adversarial Examples on NLP
 Paper: Deep Text Classification Can be Fooled, Bin Liang et al,
International Joint Conference on Artificial Intelligence, 2018.
 The character-level DNN is trained on a DBpedia ontology dataset,
which contains 560,000 training samples and 70,000 testing
samples of 14 high-level classes, such as Company, Building, Film
and so on.
34
Adversarial Examples on NLP
 Paper: Deep Text Classification Can be Fooled, Bin Liang et al,
International Joint Conference on Artificial Intelligence, 2018.
 The MR dataset is a movie review repository (containing 10,662
reviews) while CR contains 3,775 reviews about products, e.g. a
music player. Reviews from both datasets can be categorized as
either Positive or Negative.
35
Related papers
 Generating Natural Language Adversarial Examples, Moustafa et
al,. EMNLP 2018.
 TEXTBUGGER: Generating Adversarial Text Against Real-world
Applications, NDSS 2019
36
Adversarial Examples on Speech Processing
37
Introduction
Personal assistants such as Alexa, Siri, or Cortana are widely deployed
these days. Such Automatic Speech Recognition (ASR) systems can
translate and even recognize spoken language and provide a written
transcript of the spoken language. Recent advances in the fields of
deep learning and big data analysis supported significant progress for
ASR systems and have become almost as good at it as human
listeners.
What is Automatic Speech Recognition (ASR) system?
38
Automatic Speech Recognition (ASR)
Paper: Adversarial Attacks Against Automatic Speech
Recognition Systems via Psychoacoustic Hiding, Lea et al.,
NDSS 2019
39
How can we attack ASR systems?
Psychoacoustics: Human hearing is limited to a certain range
of frequencies, amplitudes, and signal dynamics. The field
of psychoacoustics addresses such hearing restrictions and
provides a rule set for things that humans can and cannot
hear. While these rules are used in different fields, e.g.,
in MP3 music compression, we can also utilize
psychoacoustics for our attack to hide noise in such a way
that humans (almost) cannot hear it.
40
Psychoacoustics
Hearing Models
Psychoacoustic hearing thresholds describe masking effects
in human acoustic perception. Probably the best-known
example for this is MP3 compression, where the
compression algorithm uses a set of computed hearing
thresholds to find out which parts of the input signal are
imperceptible for human listeners. By removing those
parts, the audio signal can be transformed into a smaller
but lossy representation, which requires less disk space and
less bandwidth to store or transmit.
41
Attack
For the attack, in principle, we use the very the same algorithm as for
the training of neural networks. The algorithm is based on gradient
descent, but instead of updating the parameters of the neural
network as for training, the audio signal is modified. We use the
hearing thresholds to avoid changes in easily perceptible parts of the
audio signal.
https://adversarial-attacks.net/
Kaldi Speech Recognition Toolkit
https://github.com/kaldi-asr/kaldi
42
Attack Model
HMM: Hidden
Markov Model
43
Attack Model
Forced alignment refers to the process by which
orthographic transcriptions are aligned to audio
recordings to automatically generate phone level
segmentation.
The main difference between “original audio” and
“raw audio” is that the original audio does not
change during the run-time of the algorithm, but
the raw audio is updated iteratively in order to
result in an adversarial example.
44
Results
Original Audio:
THE WALLOP PROPOSAL WOULD COST FIVE POINT FOUR TWO
BILLION DOLLARS OVER FIVE YEARS.
Adversarial Audio:
I BELIEVE ALL PEOPLE ARE GOOD
Noise:
45
Related papers
Imperceptible, Robust, and Targeted Adversarial
Examples for Automatic Speech Recognition, Yao
Qin et al., ICML 2019.
Robust Audio Adversarial Example for a Physical
Attack, Hiromu Yakura and Jun Sakuma, IJCAI 2019
46
Back to AE in Computer Vision
There is no way to protect AI model from
Adversarial Examples?
47
Back to AE in Computer Vision
Good news from CVPR 2017….
Paper: NO Need to Worry about Adversarial Examples in
Object Detection in Autonomous Vehicles, Jiajun Lu et al.,
CVPR 2017.
48
Back to AE in Computer Vision
But ….
Paper: Obfuscated Gradients Give a False Sense of Security:
Circumventing Defenses to Adversarial Examples, Athalye
et al., ICML 2018 (Best Paper).
49
Demonstration
• I will make demonstration of adversarial
examples on image classification tasks
by using pytorch and tensorflow with
several state-of-the-art AI models
(Inception V3-4, Resnet-152 …)
50
Future work
 In this talk I cover about adversarial attacks on image
classification, natural language processing and speech
recognition.
 Adversarial examples are raising as a very crucial problem in
deep learning.
 Defense strategies are not included in this talk, hope I have
another chance to discuss about it. 
Q & A
51
Supplements
Adversarial examples github:
https://github.com/tensorflow/cleverhans
https://github.com/IBM/adversarial-
robustness-toolbox
52
ご清聴ありがとうございました

More Related Content

Similar to A Survey on Security and Privacy of Machine Learning

Machine Learning ICS 273A
Machine Learning ICS 273AMachine Learning ICS 273A
Machine Learning ICS 273Abutest
 
Machine Learning ICS 273A
Machine Learning ICS 273AMachine Learning ICS 273A
Machine Learning ICS 273Abutest
 
A Friendly Introduction to Machine Learning
A Friendly Introduction to Machine LearningA Friendly Introduction to Machine Learning
A Friendly Introduction to Machine LearningHaptik
 
Introduction to the Artificial Intelligence and Computer Vision revolution
Introduction to the Artificial Intelligence and Computer Vision revolutionIntroduction to the Artificial Intelligence and Computer Vision revolution
Introduction to the Artificial Intelligence and Computer Vision revolutionDarian Frajberg
 
Security in the age of Artificial Intelligence
Security in the age of Artificial IntelligenceSecurity in the age of Artificial Intelligence
Security in the age of Artificial IntelligenceFaction XYZ
 
Keepler Data Tech | Entendiendo tus propios modelos predictivos
Keepler Data Tech | Entendiendo tus propios modelos predictivosKeepler Data Tech | Entendiendo tus propios modelos predictivos
Keepler Data Tech | Entendiendo tus propios modelos predictivosKeepler Data Tech
 
AI Cybersecurity: Pros & Cons. AI is reshaping cybersecurity
AI Cybersecurity: Pros & Cons. AI is reshaping cybersecurityAI Cybersecurity: Pros & Cons. AI is reshaping cybersecurity
AI Cybersecurity: Pros & Cons. AI is reshaping cybersecurityTasnim Alasali
 
Security in Machine Learning
Security in Machine LearningSecurity in Machine Learning
Security in Machine LearningFlavio Clesio
 
Jesús Seijas - Fooling Computer Vision [rooted2018]
Jesús Seijas - Fooling Computer Vision [rooted2018]Jesús Seijas - Fooling Computer Vision [rooted2018]
Jesús Seijas - Fooling Computer Vision [rooted2018]RootedCON
 
Adversarial Attacks on A.I. Systems — NextCon, Jan 2019
Adversarial Attacks on A.I. Systems — NextCon, Jan 2019Adversarial Attacks on A.I. Systems — NextCon, Jan 2019
Adversarial Attacks on A.I. Systems — NextCon, Jan 2019anant90
 
System for Detecting Deepfake in Videos – A Survey
System for Detecting Deepfake in Videos – A SurveySystem for Detecting Deepfake in Videos – A Survey
System for Detecting Deepfake in Videos – A SurveyIRJET Journal
 
Deep Learning - Hype, Reality and Applications in Manufacturing
Deep Learning - Hype, Reality and Applications in ManufacturingDeep Learning - Hype, Reality and Applications in Manufacturing
Deep Learning - Hype, Reality and Applications in ManufacturingAdam Cook
 
WILD PATTERNS - Introduction to Adversarial Machine Learning - ITASEC 2019
WILD PATTERNS - Introduction to Adversarial Machine Learning - ITASEC 2019WILD PATTERNS - Introduction to Adversarial Machine Learning - ITASEC 2019
WILD PATTERNS - Introduction to Adversarial Machine Learning - ITASEC 2019Pluribus One
 
Synergy of Human and Artificial Intelligence in Software Engineering
Synergy of Human and Artificial Intelligence in Software EngineeringSynergy of Human and Artificial Intelligence in Software Engineering
Synergy of Human and Artificial Intelligence in Software EngineeringTao Xie
 
What is ATT&CK coverage, anyway? Breadth and depth analysis with Atomic Red Team
What is ATT&CK coverage, anyway? Breadth and depth analysis with Atomic Red TeamWhat is ATT&CK coverage, anyway? Breadth and depth analysis with Atomic Red Team
What is ATT&CK coverage, anyway? Breadth and depth analysis with Atomic Red TeamMITRE ATT&CK
 
“Introducing Machine Learning and How to Teach Machines to See,” a Presentati...
“Introducing Machine Learning and How to Teach Machines to See,” a Presentati...“Introducing Machine Learning and How to Teach Machines to See,” a Presentati...
“Introducing Machine Learning and How to Teach Machines to See,” a Presentati...Edge AI and Vision Alliance
 
Machine learning for bestt group - 20170714
Machine learning for bestt group - 20170714Machine learning for bestt group - 20170714
Machine learning for bestt group - 20170714IBM Thailand Co Ltd
 
20101017 program analysis_for_security_livshits_lecture03_security
20101017 program analysis_for_security_livshits_lecture03_security20101017 program analysis_for_security_livshits_lecture03_security
20101017 program analysis_for_security_livshits_lecture03_securityComputer Science Club
 
AI - Media Art. 인공지능과 미디어아트
AI - Media Art. 인공지능과 미디어아트AI - Media Art. 인공지능과 미디어아트
AI - Media Art. 인공지능과 미디어아트Tae wook kang
 

Similar to A Survey on Security and Privacy of Machine Learning (20)

Debugging AI
Debugging AIDebugging AI
Debugging AI
 
Machine Learning ICS 273A
Machine Learning ICS 273AMachine Learning ICS 273A
Machine Learning ICS 273A
 
Machine Learning ICS 273A
Machine Learning ICS 273AMachine Learning ICS 273A
Machine Learning ICS 273A
 
A Friendly Introduction to Machine Learning
A Friendly Introduction to Machine LearningA Friendly Introduction to Machine Learning
A Friendly Introduction to Machine Learning
 
Introduction to the Artificial Intelligence and Computer Vision revolution
Introduction to the Artificial Intelligence and Computer Vision revolutionIntroduction to the Artificial Intelligence and Computer Vision revolution
Introduction to the Artificial Intelligence and Computer Vision revolution
 
Security in the age of Artificial Intelligence
Security in the age of Artificial IntelligenceSecurity in the age of Artificial Intelligence
Security in the age of Artificial Intelligence
 
Keepler Data Tech | Entendiendo tus propios modelos predictivos
Keepler Data Tech | Entendiendo tus propios modelos predictivosKeepler Data Tech | Entendiendo tus propios modelos predictivos
Keepler Data Tech | Entendiendo tus propios modelos predictivos
 
AI Cybersecurity: Pros & Cons. AI is reshaping cybersecurity
AI Cybersecurity: Pros & Cons. AI is reshaping cybersecurityAI Cybersecurity: Pros & Cons. AI is reshaping cybersecurity
AI Cybersecurity: Pros & Cons. AI is reshaping cybersecurity
 
Security in Machine Learning
Security in Machine LearningSecurity in Machine Learning
Security in Machine Learning
 
Jesús Seijas - Fooling Computer Vision [rooted2018]
Jesús Seijas - Fooling Computer Vision [rooted2018]Jesús Seijas - Fooling Computer Vision [rooted2018]
Jesús Seijas - Fooling Computer Vision [rooted2018]
 
Adversarial Attacks on A.I. Systems — NextCon, Jan 2019
Adversarial Attacks on A.I. Systems — NextCon, Jan 2019Adversarial Attacks on A.I. Systems — NextCon, Jan 2019
Adversarial Attacks on A.I. Systems — NextCon, Jan 2019
 
System for Detecting Deepfake in Videos – A Survey
System for Detecting Deepfake in Videos – A SurveySystem for Detecting Deepfake in Videos – A Survey
System for Detecting Deepfake in Videos – A Survey
 
Deep Learning - Hype, Reality and Applications in Manufacturing
Deep Learning - Hype, Reality and Applications in ManufacturingDeep Learning - Hype, Reality and Applications in Manufacturing
Deep Learning - Hype, Reality and Applications in Manufacturing
 
WILD PATTERNS - Introduction to Adversarial Machine Learning - ITASEC 2019
WILD PATTERNS - Introduction to Adversarial Machine Learning - ITASEC 2019WILD PATTERNS - Introduction to Adversarial Machine Learning - ITASEC 2019
WILD PATTERNS - Introduction to Adversarial Machine Learning - ITASEC 2019
 
Synergy of Human and Artificial Intelligence in Software Engineering
Synergy of Human and Artificial Intelligence in Software EngineeringSynergy of Human and Artificial Intelligence in Software Engineering
Synergy of Human and Artificial Intelligence in Software Engineering
 
What is ATT&CK coverage, anyway? Breadth and depth analysis with Atomic Red Team
What is ATT&CK coverage, anyway? Breadth and depth analysis with Atomic Red TeamWhat is ATT&CK coverage, anyway? Breadth and depth analysis with Atomic Red Team
What is ATT&CK coverage, anyway? Breadth and depth analysis with Atomic Red Team
 
“Introducing Machine Learning and How to Teach Machines to See,” a Presentati...
“Introducing Machine Learning and How to Teach Machines to See,” a Presentati...“Introducing Machine Learning and How to Teach Machines to See,” a Presentati...
“Introducing Machine Learning and How to Teach Machines to See,” a Presentati...
 
Machine learning for bestt group - 20170714
Machine learning for bestt group - 20170714Machine learning for bestt group - 20170714
Machine learning for bestt group - 20170714
 
20101017 program analysis_for_security_livshits_lecture03_security
20101017 program analysis_for_security_livshits_lecture03_security20101017 program analysis_for_security_livshits_lecture03_security
20101017 program analysis_for_security_livshits_lecture03_security
 
AI - Media Art. 인공지능과 미디어아트
AI - Media Art. 인공지능과 미디어아트AI - Media Art. 인공지능과 미디어아트
AI - Media Art. 인공지능과 미디어아트
 

Recently uploaded

VIRUSES structure and classification ppt by Dr.Prince C P
VIRUSES structure and classification ppt by Dr.Prince C PVIRUSES structure and classification ppt by Dr.Prince C P
VIRUSES structure and classification ppt by Dr.Prince C PPRINCE C P
 
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...Sérgio Sacani
 
STERILITY TESTING OF PHARMACEUTICALS ppt by DR.C.P.PRINCE
STERILITY TESTING OF PHARMACEUTICALS ppt by DR.C.P.PRINCESTERILITY TESTING OF PHARMACEUTICALS ppt by DR.C.P.PRINCE
STERILITY TESTING OF PHARMACEUTICALS ppt by DR.C.P.PRINCEPRINCE C P
 
GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)Areesha Ahmad
 
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptxUnlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptxanandsmhk
 
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Lokesh Kothari
 
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...Sérgio Sacani
 
Biopesticide (2).pptx .This slides helps to know the different types of biop...
Biopesticide (2).pptx  .This slides helps to know the different types of biop...Biopesticide (2).pptx  .This slides helps to know the different types of biop...
Biopesticide (2).pptx .This slides helps to know the different types of biop...RohitNehra6
 
Orientation, design and principles of polyhouse
Orientation, design and principles of polyhouseOrientation, design and principles of polyhouse
Orientation, design and principles of polyhousejana861314
 
Recombinant DNA technology (Immunological screening)
Recombinant DNA technology (Immunological screening)Recombinant DNA technology (Immunological screening)
Recombinant DNA technology (Immunological screening)PraveenaKalaiselvan1
 
Hubble Asteroid Hunter III. Physical properties of newly found asteroids
Hubble Asteroid Hunter III. Physical properties of newly found asteroidsHubble Asteroid Hunter III. Physical properties of newly found asteroids
Hubble Asteroid Hunter III. Physical properties of newly found asteroidsSérgio Sacani
 
DIFFERENCE IN BACK CROSS AND TEST CROSS
DIFFERENCE IN  BACK CROSS AND TEST CROSSDIFFERENCE IN  BACK CROSS AND TEST CROSS
DIFFERENCE IN BACK CROSS AND TEST CROSSLeenakshiTyagi
 
Botany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdfBotany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdfSumit Kumar yadav
 
Chromatin Structure | EUCHROMATIN | HETEROCHROMATIN
Chromatin Structure | EUCHROMATIN | HETEROCHROMATINChromatin Structure | EUCHROMATIN | HETEROCHROMATIN
Chromatin Structure | EUCHROMATIN | HETEROCHROMATINsankalpkumarsahoo174
 
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral AnalysisRaman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral AnalysisDiwakar Mishra
 
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...anilsa9823
 
Disentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTDisentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTSérgio Sacani
 
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...ssifa0344
 

Recently uploaded (20)

VIRUSES structure and classification ppt by Dr.Prince C P
VIRUSES structure and classification ppt by Dr.Prince C PVIRUSES structure and classification ppt by Dr.Prince C P
VIRUSES structure and classification ppt by Dr.Prince C P
 
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
 
STERILITY TESTING OF PHARMACEUTICALS ppt by DR.C.P.PRINCE
STERILITY TESTING OF PHARMACEUTICALS ppt by DR.C.P.PRINCESTERILITY TESTING OF PHARMACEUTICALS ppt by DR.C.P.PRINCE
STERILITY TESTING OF PHARMACEUTICALS ppt by DR.C.P.PRINCE
 
GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)
 
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptxUnlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
 
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
 
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
 
Engler and Prantl system of classification in plant taxonomy
Engler and Prantl system of classification in plant taxonomyEngler and Prantl system of classification in plant taxonomy
Engler and Prantl system of classification in plant taxonomy
 
Biopesticide (2).pptx .This slides helps to know the different types of biop...
Biopesticide (2).pptx  .This slides helps to know the different types of biop...Biopesticide (2).pptx  .This slides helps to know the different types of biop...
Biopesticide (2).pptx .This slides helps to know the different types of biop...
 
Orientation, design and principles of polyhouse
Orientation, design and principles of polyhouseOrientation, design and principles of polyhouse
Orientation, design and principles of polyhouse
 
Recombinant DNA technology (Immunological screening)
Recombinant DNA technology (Immunological screening)Recombinant DNA technology (Immunological screening)
Recombinant DNA technology (Immunological screening)
 
9953056974 Young Call Girls In Mahavir enclave Indian Quality Escort service
9953056974 Young Call Girls In Mahavir enclave Indian Quality Escort service9953056974 Young Call Girls In Mahavir enclave Indian Quality Escort service
9953056974 Young Call Girls In Mahavir enclave Indian Quality Escort service
 
Hubble Asteroid Hunter III. Physical properties of newly found asteroids
Hubble Asteroid Hunter III. Physical properties of newly found asteroidsHubble Asteroid Hunter III. Physical properties of newly found asteroids
Hubble Asteroid Hunter III. Physical properties of newly found asteroids
 
DIFFERENCE IN BACK CROSS AND TEST CROSS
DIFFERENCE IN  BACK CROSS AND TEST CROSSDIFFERENCE IN  BACK CROSS AND TEST CROSS
DIFFERENCE IN BACK CROSS AND TEST CROSS
 
Botany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdfBotany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdf
 
Chromatin Structure | EUCHROMATIN | HETEROCHROMATIN
Chromatin Structure | EUCHROMATIN | HETEROCHROMATINChromatin Structure | EUCHROMATIN | HETEROCHROMATIN
Chromatin Structure | EUCHROMATIN | HETEROCHROMATIN
 
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral AnalysisRaman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
 
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
Lucknow 💋 Russian Call Girls Lucknow Finest Escorts Service 8923113531 Availa...
 
Disentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTDisentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOST
 
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
 

A Survey on Security and Privacy of Machine Learning

  • 1. 1 A Survey on Security and Privacy of Machine Learning Dang Duy Thang September 15th 2019
  • 2. 2 Big thanks to VJAI community and BizReach company for supporting this talk!
  • 3. 3 Outline 1. Introduction 2. Adversarial examples on CV 3. Adversarial examples on NLP 4. Adversarial examples on Speech 5. Conclusion
  • 4. 4 Introduction What can AI learn from security? • History has shown attacker always follows footsteps of new technology development. • As AI controls more and more systems, attacker will have higher & higher chances.
  • 5. 5 Failures of ML Microsoft’s AI chatbot The bot was Artificial Intelligence Chatterbot that created by Microsoft, and named "Tay" after the acronym "thinking about you". There are similar platforms as Xiaoice in China and Rinna in Japan. • Tay was designed to mimic the language patterns of a 19-year-old American girl, and to learn from interacting with human users of Twitter. • The system collected vast amounts of intimate details on individuals, the program raises privacy questions.
  • 6. 6 Failures of ML Microsoft’s AI chatbot Microsoft had to shut down “TAY” only 16 hours after its launch on March 23, 2016 because it started tweeting Racist messages such as: “Tay accused George W. Bush of causing 9/11, praise Hitler, refered to President Barack Obama as a “monkey” Tay's behavior on a "coordinated attack by a subset of people" that "exploited a vulnerability in Tay.
  • 7. 7 What is the adversarial examples? In the paper “Explaining and harnessing Adversarial Examples”, Goodfellow et al, International Conference on Learning Representations 2015 : “Adversarial examples is that ML models misclassify examples that are only slightly different from correctly classified examples draw from the data distribution”.
  • 8. 8 Adversarial Examples Research fields are effected by Adversarial Examples 1. Computer vision 2. Natural language processing 3. Speech processing
  • 9. 9 Adversarial Examples is a big problem? https://nicholas.carlini.com/writing/2019 /all-adversarial-example-papers.html NeurIPS - Workshop on Security in Machine Learning 2019 ICML - Workshop on the Security and Privacy of Machine Learning 2019 ICLR - Safe Machine Learning Specification, Robustness and Assurance Workshop 2019 CVPR - The Bright and Dark Sides of Computer Vision: Challenges and Opportunities for Privacy and Security Workshop 2019 https://github.com/IBM/adversar ial-robustness-toolbox https://github.com/tensorflow/cleverhans
  • 10. 10 Adversarial Examples There are two types of adversarial examples attacks: 1. White-box attacks 2. Black-box attacks Or we can categorize AE as output type: 1. Untargeted attack 2. Targeted attack
  • 11. 11 Image Classification http://cs231n.github.io Each number is an integer that ranges from 0 (black) to 255 (white)
  • 12. 12 Interpreting a linear classifier Score function that maps the raw image pixels to class scores Higher score means better, or ('more likely') Loss function: measures quality of outcomes the loss will be high if we’re doing a poor classification, and it will be low if we’re doing well. http://cs231n.github.io
  • 13. 13 72 136 56 103 61 52 134 42 142 210 134 99 150 46 81 76 72 136 56 201 61 52 134 172 142 210 134 81 48 35 110 200 How AI system recognizes a RGB Image? Input RGB Image “vending machine” Red channel Green channel Blue channel 72 136 56 80 61 52 134 97 142 210 134 178 36 79 63 85 What a computer sees Three matrices correspond to three channels 3つの行列が3つのチャンネルに対応
  • 14. 14 72 136 56 103 61 52 134 42 142 210 134 99 150 46 81 76 72 136 56 201 61 52 134 172 142 210 134 81 48 35 110 200 How AI system recognizes a RGB Image? 72 136 56 80 61 52 134 97 142 210 134 178 36 79 63 85 Flatten 72 136 56 80 … … 201 172 81 … … 103 42 99 … 3 matrices Vector vectorization
  • 15. 15 How AI system recognizes a RGB Image? 72 136 56 … … … … … 201 … … … 103 … … Vector Input Layer Hidden Layer Output Layer thatch keyboard …
  • 16. 16 AI Model Let’s consider the Google Inception V3 that was trained on ImageNet dataset with 1,000 classes. Input image is a color image with 299x299 size. Input RGB Image 299 x 299 72 136 56 103 61 52 134 42 142 210 134 99 150 46 81 76 72 136 56 201 61 52 134 172 142 210 134 81 48 35 110 200 72 136 56 80 61 52 134 97 142 210 134 178 36 79 63 85 299 x 299 x3 1x268,203 scaling Vectorization 72 136 56 80 … … 201 172 81 … … 103 42 99 …
  • 17. 17 Normal process In a normal process to train a neural network: ニューラルネットワークを訓練する通常のプロセ スでは Fix, no change 固定、変更なし Fine-tune W respect to Loss(input, label) Try to find the best W that makes Loss function is as small as possible 損失に対するWの微調整(入力、ラベル) 損失関数ができるだけ小さくなるような最良のWを見つけようとする label http://cs231n.github.io
  • 18. 18 Crafting Adversarial Examples In a process to create adversarial examples: 敵対的な例を作成する過程で Fine-tune input respect to Loss(input, label) Keep Loss(input, label) is large Keep Loss(adversarial, new label) is small label Fix, no change 固定、変更なし 損失に対する入力の微調整(入力、ラベル) (入力、ラベル)の損失は大きく、 (敵対的、新しいラベル)の損失は小さくする http://cs231n.github.io
  • 19. 19 Adversarial Attack Input vector AI System 46 140 37 92 … … 118 37 41 … … 90 29 30 … Adversarial Vector 72 136 56 90 61 52 134 29 142 210 134 30 150 46 81 76 72 136 56 118 61 52 134 37 142 210 134 41 48 35 110 200 46 140 37 92 61 52 134 97 142 210 134 178 36 79 63 85 299 x 299 x3 72 136 56 80 … … 201 172 81 … … 103 42 99 … Adversarial Image reform
  • 20. 20 Back to FGSM (Fast Gradient Sign Method) FGSM (Fast Gradient Sign Method)*: (*) Ian Goodfellow et al, Explaining and harnessing adversarial examples, ICLR 2015. 72 136 56 80 … … 201 172 81 … … 103 42 99 … 46 140 37 92 … … 118 37 41 … … 90 29 30 … = + adversarial perturbation
  • 21. 21 Adversarial Examples is actually dangerous? Paper: Fooling automated surveillance cameras: adversarial patches to attack person detection, CVPR 2018
  • 22. 22 Adversarial Examples is actually dangerous? Autonomous car are built on ML/AI for auto detect traffic sign? What happen if that car’s AI system does misrecognition “STOP” traffic sign instead of “turn right” sign when it is running in a high way? Taesoo Kim, AI & Security, Microsoft Research Faculty Summit 2017
  • 23. 23 However, white-box attack is not really realistic because you’re unlikely to get access to the gradients of the loss function on any particular system. -> The paper from Papernot et al, “Practical Black- Box Attacks against Machine Learning”, ASIA CCS 2017. Adversarial Samples – Black-box attacks
  • 24. 24  In a more realistic context, you would want to attack a system having only access to its outputs.  In the paper “Practical Black-Box Attacks against Machine Learning” proposed a substitute Deep Neural Networks to solve the same classification tasks as the target model. Adversarial Samples – Black-box attacks
  • 25. 25 There are two main strategies in black-box attacks: 1. Create a synthetic dataset. 2. Build and Train a local substitute Deep Neural Networks. Adversarial Samples – Black-box attacks
  • 26. 26 How to create a synthetic dataset (input, output)? 1. Inputs: are synthetic and generated by the adversary. 2. Outputs: are labels assigned by the target DNN and observed by the adversary. Adversarial Samples – Black-box attacks
  • 27. 27 How to build a local substitute Deep Neural Networks? The attacker queries the oracle (target system) with synthetic inputs selected by a Jacobian-based heuristic to build a model F approximating the oracle model O’s decision boundaries. Adversarial Samples – Black-box attacks
  • 28. 28 What are the Oracle DNNs?: 1. Amazon oracle: https://aws.amazon.com/machine- learning 2. Google oracle: https://cloud.google.com/prediction 3. MetaMind oracle: https://einstein.ai/ They provide some functionalities: dataset upload, automated model training and model prediction querying. This method made MetaMind misclassify at 84.24%, Amazon and Google at 96.19% and 88.94%. Adversarial Samples – Black-box attacks
  • 29. 29 In an another real scenario, the attacker would not be allowed to provide its own image files, the neural network would take camera pictures as input. That’s the problem, the authors of this paper “Adversarial examples in the physical world”(ICLR, 2017) are trying to solve. Adversarial Samples – Black-box attacks
  • 30. 30 “We used images taken from a cell-phone camera as a input to an image classification neural network” Adversarial Samples – Black-box attacks
  • 31. 31 Adversarial Examples in Natural Language Processing
  • 32. 32 Adversarial Examples on NLP  Paper: Deep Text Classification Can be Fooled, Bin Liang et al, International Joint Conference on Artificial Intelligence, 2018.  The foundation of attack lies in identifying the text items that possess significant contribution to the classification by leveraging the cost gradient.  They determine the most frequent phrases, called Hot Training Phrases (HTPs).
  • 33. 33 Adversarial Examples on NLP  Paper: Deep Text Classification Can be Fooled, Bin Liang et al, International Joint Conference on Artificial Intelligence, 2018.  The character-level DNN is trained on a DBpedia ontology dataset, which contains 560,000 training samples and 70,000 testing samples of 14 high-level classes, such as Company, Building, Film and so on.
  • 34. 34 Adversarial Examples on NLP  Paper: Deep Text Classification Can be Fooled, Bin Liang et al, International Joint Conference on Artificial Intelligence, 2018.  The MR dataset is a movie review repository (containing 10,662 reviews) while CR contains 3,775 reviews about products, e.g. a music player. Reviews from both datasets can be categorized as either Positive or Negative.
  • 35. 35 Related papers  Generating Natural Language Adversarial Examples, Moustafa et al,. EMNLP 2018.  TEXTBUGGER: Generating Adversarial Text Against Real-world Applications, NDSS 2019
  • 36. 36 Adversarial Examples on Speech Processing
  • 37. 37 Introduction Personal assistants such as Alexa, Siri, or Cortana are widely deployed these days. Such Automatic Speech Recognition (ASR) systems can translate and even recognize spoken language and provide a written transcript of the spoken language. Recent advances in the fields of deep learning and big data analysis supported significant progress for ASR systems and have become almost as good at it as human listeners. What is Automatic Speech Recognition (ASR) system?
  • 38. 38 Automatic Speech Recognition (ASR) Paper: Adversarial Attacks Against Automatic Speech Recognition Systems via Psychoacoustic Hiding, Lea et al., NDSS 2019
  • 39. 39 How can we attack ASR systems? Psychoacoustics: Human hearing is limited to a certain range of frequencies, amplitudes, and signal dynamics. The field of psychoacoustics addresses such hearing restrictions and provides a rule set for things that humans can and cannot hear. While these rules are used in different fields, e.g., in MP3 music compression, we can also utilize psychoacoustics for our attack to hide noise in such a way that humans (almost) cannot hear it.
  • 40. 40 Psychoacoustics Hearing Models Psychoacoustic hearing thresholds describe masking effects in human acoustic perception. Probably the best-known example for this is MP3 compression, where the compression algorithm uses a set of computed hearing thresholds to find out which parts of the input signal are imperceptible for human listeners. By removing those parts, the audio signal can be transformed into a smaller but lossy representation, which requires less disk space and less bandwidth to store or transmit.
  • 41. 41 Attack For the attack, in principle, we use the very the same algorithm as for the training of neural networks. The algorithm is based on gradient descent, but instead of updating the parameters of the neural network as for training, the audio signal is modified. We use the hearing thresholds to avoid changes in easily perceptible parts of the audio signal. https://adversarial-attacks.net/ Kaldi Speech Recognition Toolkit https://github.com/kaldi-asr/kaldi
  • 43. 43 Attack Model Forced alignment refers to the process by which orthographic transcriptions are aligned to audio recordings to automatically generate phone level segmentation. The main difference between “original audio” and “raw audio” is that the original audio does not change during the run-time of the algorithm, but the raw audio is updated iteratively in order to result in an adversarial example.
  • 44. 44 Results Original Audio: THE WALLOP PROPOSAL WOULD COST FIVE POINT FOUR TWO BILLION DOLLARS OVER FIVE YEARS. Adversarial Audio: I BELIEVE ALL PEOPLE ARE GOOD Noise:
  • 45. 45 Related papers Imperceptible, Robust, and Targeted Adversarial Examples for Automatic Speech Recognition, Yao Qin et al., ICML 2019. Robust Audio Adversarial Example for a Physical Attack, Hiromu Yakura and Jun Sakuma, IJCAI 2019
  • 46. 46 Back to AE in Computer Vision There is no way to protect AI model from Adversarial Examples?
  • 47. 47 Back to AE in Computer Vision Good news from CVPR 2017…. Paper: NO Need to Worry about Adversarial Examples in Object Detection in Autonomous Vehicles, Jiajun Lu et al., CVPR 2017.
  • 48. 48 Back to AE in Computer Vision But …. Paper: Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples, Athalye et al., ICML 2018 (Best Paper).
  • 49. 49 Demonstration • I will make demonstration of adversarial examples on image classification tasks by using pytorch and tensorflow with several state-of-the-art AI models (Inception V3-4, Resnet-152 …)
  • 50. 50 Future work  In this talk I cover about adversarial attacks on image classification, natural language processing and speech recognition.  Adversarial examples are raising as a very crucial problem in deep learning.  Defense strategies are not included in this talk, hope I have another chance to discuss about it.  Q & A

Editor's Notes

  1. Artificial intelligence and machine learning are being applied more broadly across different industries and applications than ever before, and cyber security is no exception.  In a cybersecurity context, AI is software that perceives its environment well enough to identify events and take action against a predefined purpose. AI is particularly good at recognizing patterns and anomalies within them, which makes it an excellent tool to detect threats. Machine learning algorithms can be used to create profiles of normal behavior, and these profiles can be either more global, or alternatively either user or host based. Based on these, it is possible to differentiate normal and abnormal behavior practically in real time.
  2. For example, in the image below an image classification model takes a single image and assigns probabilities to 4 labels, {cat, dog, hat, mug}. As shown in the image, keep in mind that to a computer an image is represented as one large 3-dimensional array of numbers. In this example, the cat image is 299 pixels wide, 299 pixels tall, and has three color channels Red,Green,Blue (or RGB for short). Therefore, the image consists of 299 x 299 x 3 numbers, or a total of 268,203 numbers. Each number is an integer that ranges from 0 (black) to 255 (white). Our task is to turn this quarter of a million numbers into a single label, such as “cat”.
  3. An example of mapping an image to class scores. For the sake of visualization, we assume the image only has 4 pixels (4 monochrome pixels, we are not considering color channels in this example for brevity), and that we have 3 classes (red (cat), green (dog), blue (ship) class). (Clarification: in particular, the colors here simply indicate 3 classes and are not related to the RGB channels.) We stretch the image pixels into a column and perform matrix multiplication to get the scores for each class. Note that this particular set of weights W is not good at all: the weights assign our cat image a very low cat score. In particular, this set of weights seems convinced that it's looking at a dog.
  4. ASIA CCS : Proceedings of the 2017 ACM on Asia Conference on Computer and Communications Security. Papernot et al proposed a way to deal with black-box attack.
  5. The problem with this is that you would not be able to apply the FGSM algorithm anymore as you would not have access to the network itself.
  6. In this paper, we introduce the first demonstration that black-box attacks against DNN classifiers are practical for real-world adversaries with no knowledge about the model. We assume the adversary (a) has no information about the structure or parameters of the DNN, and (b) does not have access to any large training dataset. The adversary’s only capability is to observe labels assigned by the DNN for chosen inputs, in a manner analog to a cryptographic oracle. MNIST comprises 60.000 training and 10.000 test images of handwritten digits. The task associated with the dataset is to identify the digit corresponding to each image. Each 28x28 grayscale sample is encoded as a vector of pixel intensities in the interval The GTSRB dataset is an image collection consisting of 43 traffic signs [13]. Images vary in size and are RGB-encoded. To simplify, we resize images to 32x32 pixels, re center them by subtracting the mean component, and rescale them by factoring their standard deviations out. We keep 35.000 images for our training set and 4.000 for our validation set (out of the 39.209 available), and 10.000 for our test set (out of 12.630).
  7. “We used images taken from a cell-phone camera as a input to an Inception v3 image classification neural network. We showed that in such a set-up, a significant fraction of adversarial images crafted using the original network are misclassified even when fed to the classifier through the camera.” All previous work has assumed a threat model in which the adversary can feed data directly into the machine learning classifier. This is not always the case for systems operating in the physical world, for example those which are using signals from cameras and other sensors as input. This paper shows that even in such physical world scenarios, machine learning systems are vulnerable to adversarial examples. We demonstrate this by feeding adversarial images obtained from a cell-phone camera to an ImageNet Inception classifier and measuring the classification accuracy of the system. We find that a large fraction of adversarial examples are classified incorrectly even when perceived through the camera
  8. Personal assistants such as Alexa, Siri, or Cortana are widely deployed these days. Such Automatic Speech Recognition (ASR) systems can translate and even recognize spoken language and provide a written transcript of the spoken language. Recent advances in the fields of deep learning and big data analysis supported significant progress for ASR systems and have become almost as good at it as human listeners. 
  9. Personal assistants such as Alexa, Siri, or Cortana are widely deployed these days. Such Automatic Speech Recognition (ASR) systems can translate and even recognize spoken language and provide a written transcript of the spoken language. Recent advances in the fields of deep learning and big data analysis supported significant progress for ASR systems and have become almost as good at it as human listeners. 
  10. Psychoacoustics: Human hearing is limited to a certain range of frequencies, amplitudes, and signal dynamics. The field of psychoacoustics addresses such hearing restrictions and provides a rule set for things that humans can and cannot hear. While these rules are used in different fields, e.g., in MP3 music compression, we can also utilize psychoacoustics for our attack to hide noise in such a way that humans (almost) cannot hear it.
  11. Hearing Models Psychoacoustic hearing thresholds describe masking effects in human acoustic perception. Probably the best-known example for this is MP3 compression, where the compression algorithm uses a set of computed hearing thresholds to find out which parts of the input signal are imperceptible for human listeners. By removing those parts, the audio signal can be transformed into a smaller but lossy representation, which requires less disk space and less bandwidth to store or transmit.
  12. For the attack, in principle, we use the very the same algorithm as for the training of neural networks. The algorithm is based on gradient descent, but instead of updating the parameters of the neural network as for training, the audio signal is modified. We use the hearing thresholds to avoid changes in easily perceptible parts of the audio signal.
  13. The main difference between original audio and raw audio is that the original audio does not change during the run-time of the algorithm, but the raw audio is updated iteratively in order to result in an adversarial example. The algorithm uses the original audio signal and the target transcription as inputs in order to find the best target pseudo-posteriors. The forced alignment is performed once at the beginning of the algorithm. The hearing thresholds are applied during the backpropagation in order to limit the changes that are perceptible by a human. One major problem of attacks against ASR systems is that they require the recognition to pass through a certain sequence of HMM states in such a way that it leads to the target transcription. However, due to the decoding step— which includes a graph search—for a given transcription, many valid pseudo-posterior combinations exist. For example, when the same text is spoken at different speeds, the sequence of the HMM states is correspondingly faster or slower. We can benefit from this fact by using that version of pseudo-posteriors which best fits the given audio signal and the desired target transcription. We use forced alignment as an algorithm for finding the best possible temporal alignment between the acoustic signal that we manipulate and the transcription that we wish to obtain. This algorithm is provided by the Kaldi toolkit. Note that it is not always possible to find an alignment that fits an audio file to any target transcription. In this case, we set the alignment by dividing the audio sample equally into the number of states and set the target according to this division.
  14. Forced alignment refers to the process by which orthographic transcriptions are aligned to audio recordings to automatically generate phone level segmentation. The main difference between “original audio” and “raw audio” is that the original audio does not change during the run-time of the algorithm, but the raw audio is updated iteratively in order to result in an adversarial example.
  15. For the attack, in principle, we use the very the same algorithm as for the training of neural networks. The algorithm is based on gradient descent, but instead of updating the parameters of the neural network as for training, the audio signal is modified. We use the hearing thresholds to avoid changes in easily perceptible parts of the audio signal.
  16. NO Need to Worry about Adversarial Examples in Object Detection in Autonomous Vehicles, CVPR 2017 Synthesizing Robust Adversarial Examples, ICML 2018