A Tour of Machine Learning Security Florian Tramèr CISPA August 6 th 2018
The Deep Learning Revolution First they came for images…
The Deep Learning Revolution And then everything else…
The ML Revolution Including things that likely won’t work… 4
What does this mean for privacy & security? Crypto, Trusted hardware Differential privacy Privacy & integrity Data inference Outsourced learning Model theft Test outputs Training data Data poisoning Blockchain ??? dog cat bird Robust statistics Adversarial examples Outsourced inference Test data Privacy & integrity Crypto, Trusted hardware Adapted from (Goodfellow 2018) 5
This talk: security of deployed models Crypto, Trusted hardware Differential privacy Privacy & integrity Data inference Outsourced learning Model theft Test outputs Training data Data poisoning ??? dog cat bird Robust statistics Adversarial examples Outsourced inference Test data Privacy & integrity Crypto, Trusted hardware 6
Stealing ML Models Crypto, Trusted hardware Differential privacy Privacy & integrity Data inference Outsourced learning Model theft Test outputs Training data Data poisoning ??? dog cat bird Robust statistics Adversarial examples Outsourced inference Test data Privacy & integrity Crypto, Trusted hardware 7
Machine Learning as a Service Goal 2: Model Confidentiality • Model/Data Monetization • Sensitive Data Model f Training API Prediction API input classification Data Black Box Goal 1: Rich Prediction APIs $$$ per query • Highly Available • High-Precision Results 8
Model Extraction Goal: Adversarial client learns close approximation of f using as few queries as possible x Data Attack Model f f’ f(x) Applications: 1) Undermine pay-for-prediction pricing model 2) ”White-box” attacks: › Infer private training data › Model evasion (adversarial examples) 9
Model Extraction Goal: Adversarial client learns close approximation of f using as few queries as possible x Data Attack Model f f’ f(x) Isn’t this “just No! Prediction APIs Machine Learning”? return fine-grained information that makes extracting much easier than learning 10
Learning vs Extraction Learning f(x) Extracting f(x) Function to learn Noisy real-world “Simple” deterministic function f(x) phenomenon 11
Learning vs Extraction Learning f(x) Extracting f(x) Function to learn Noisy real-world “Simple” deterministic function f(x) phenomenon Available labels hard labels Depending on API: (e.g., “cat”, “dog”, …) - Hard labels - Soft labels (class probas) - Gradients (for interpretability) 12
Learning vs Extraction Learning f(x) Extracting f(x) Function to learn Noisy real-world “Simple” deterministic function f(x) phenomenon Available labels hard labels Depending on API: (e.g., “cat”, “dog”, …) - Hard labels - Soft labels (class probas) - Gradients (Milli et al.) Labeling function Humans, real-world Query f(x) on any input x data collection => No need for labeled data => Queries can be adaptive 13
Learning vs Extraction for specific models Learning f(x) Extracting f(x) Logistic |Data| ≈ 10 * |Features| - Hard labels only: (Loyd & Meek) Regression - With confidences: simple system of equations ( T et al.) |Data| = |Features| + cte 14
Learning vs Extraction for specific models Learning f(x) Extracting f(x) Logistic |Data| ≈ 10 * |Features| - Hard labels only: (Loyd & Meek) Regression - With confidences: simple system of equations ( T et al.) |Data| = |Features| + cte Decision - NP-hard in general “Differential testing” algorithm to Trees - polytime for Boolean trees recover the full tree ( T et al.) (Kushilevitz & Mansour) 15
Learning vs Extraction for specific models Learning f(x) Extracting f(x) Logistic |Data| ≈ 10 * |Features| - Hard labels only: (Loyd & Meek) Regression - With confidences: simple system of equations ( T et al.) |Data| = |Features| + cte Decision - NP-hard in general “Differential testing” algorithm to Trees - polytime for Boolean trees recover the full tree ( T et al.) (Kushilevitz & Mansour) Neural Large models required - Distillation (Hinton et al.) Networks “The more data the better” Make smaller copy of model from confidence scores - Extraction from hard labels No quantitative analysis for (Papernot et al., T et al.) large neural nets yet 16
Crypto, Trusted hardware Differential privacy Privacy & integrity Data inference Outsourced learning Model theft Test outputs Training data Data poisoning ??? dog cat bird Robust statistics Adversarial examples Outsourced inference Test data Privacy & integrity Crypto, Trusted hardware 17
Trusted execution of ML: 3 motivating scenarios 1. Cloud ML APIs Data Privacy Integrity - Model “downgrade” - Disparate impact - Other malicious tampering 18
Trusted execution of ML: 3 motivating scenarios 2. Federated Learning Integrity Data privacy Poison model updates 19
Trusted execution of ML: 3 motivating scenarios 3. Trojaned hardware (Verifiable ASICs model, Wahby et al.) Integrity 20
Solutions Cryptography / Statistics 1. Outsourced ML : FHE, MPC, (ZK) proof systems 2. Federated learning : robust statistics? 3. Trojaned hardware : some root of trust is needed Trusted Execution Environments (TEEs) 1. Outsourced ML : secure enclaves 2. Federated learning : trusted sensors + secure enclaves 3. Trojaned hardware: fully trusted (but possibly slow) hardware
Trusted Execution: At what cost? • Trusted ASICs (Wahby et al.): ~10 8 � worse than SOTA • Intel SGX: VGG16 Inference 400 350 350 300 Paging at Images / sec 250 ~90MB 200 150 100 50 1 0 GPU SGX GPU: Nvidia TITAN XP SGX: Intel Core i7-6700 Skylake Single Core @ 3.40GHz https://medium.com/@danny_harnik/impressions-of-intel-sgx-performance-22442093595a
“How do we efficiently leverage TEEs for secure machine learning computations?” Idea: outsource work to collocated , faster but untrusted device and verify results x TEE F(x), proof Computations Gap between trusted Privacy and untrusted device Verifiable ASICs Arithmetic circuits ~ 8 orders of magnitude No (Wahby et al., 2016) (GKR protocol) Slalom DNN inference ~ 1-2 orders “Yes” Not in this talk
Bottlenecks in deep neural networks non linear stuff (cheap) MATRIX MULTIPLICATION ~ 97% VGG16 Inference on 1 CPU core
Outsourcing matrix multiplication: Freivald’s algorithm X ∈ " n ⨉ n , W ∈ " n ⨉ n Input: DNN weights. Fixed at inference time Direct Compute: Z = X * W ≈ n 3 multiplications or O(n 2.81 ) with Strassen Outsource + Verify: • Sample r ← " n uniformly at random • Check: Z * r = X * (W * r) • Complexity: ≈ 3n 2 multiplications • Soundness: 1 / | " | (boost by repeating)
Freivald variants for arbitrary linear operators Linear operator: z = F(x) = x * A Matrix of size |x| × |z| Vector of size |z| Vector of size |x| • Batched verification Z = F ( [x 1 … x B ] ) = [ F(x 1 ) … F(x B ) ] ⇒ Complexity = B*cost(F) Freivald: r * Z = F ( r * [x 1 … x B ] ) ⇒ Complexity = B*(|x|+|z|) + cost(F) • With precomputation Precompute A’ = A * r = ( ∇ x F)(r) Freivald: z * r = x * A’ ⇒ Complexity = |x|+|z| 2 inner products!
Slalom Summary Slalom TEE X 1 Z 1 = X 1 * W 1 1. Freivald check for (X 1 , W 1 , Z 1 ) 2. X 2 = σ(Z 1 ) X 2 Z 2 = X 2 * W 2 Arbitrary non-linearity … 27
Design and Evaluation TEE • TEE: Intel SGX ”Desktop” CPU (single thread) • Untrusted device: Nvidia Tesla GPU • Port of the Eigen linear algebra C++ library to SGX (used in e.g., TensorFlow) • No simulation mode! VGG16 MobileNet 25 120 97.1 19.6 100 20 Images / sec 80 15 60 10 40 30 15.9 5 20 1.7 1 0 0 Compute Verify Verify with Compute Verify Verify with preproc preproc 28
Crypto, Trusted hardware Differential privacy Privacy & integrity Data inference Outsourced learning Model theft Test outputs Training data Data poisoning ??? dog cat bird Robust statistics Adversarial examples Outsourced inference Test data Privacy & integrity Crypto, Trusted hardware 29
ML models make surprising mistakes + . 007 ⇥ = Pretty sure this I’m certain this is a panda is a gibbon (Szegedy et al. 2013, Goodfellow et al. 2015) 30
Where are the defenses? • Adversarial training Prevent “all/most Szegedy et al. 2013, Goodfellow et al. 2015, attacks” for a Kurakin et al. 2016, T et al. 2017, Madry et al. 2017, Kannan et al. 2018 given norm ball • Convex relaxations with provable guarantees Raghunathan et al. 2018, Kolter & Wong 2018, Sinha et al. 2018 • A lot of broken defenses… 31
Do we have a realistic threat model? (no…) Current approach: 1. Fix a ”toy” attack model (e.g., some l ∞ ball) 2. Directly optimize over the robustness measure Þ Defenses do not generalize to other attack models Þ Defenses are meaningless for applied security What do we want? • Model is “always correct” (sure, why not?) • Model has blind spots that are “hard to find” • “Non-information-theoretic” notions of robustness? • CAPTCHA threat model is interesting to think about 32
Recommend
More recommend
