ترغب بنشر مسار تعليمي؟ اضغط هنا

Reward Poisoning in Reinforcement Learning: Attacks Against Unknown Learners in Unknown Environments

187   0   0.0 ( 0 )
 نشر من قبل Adish Singla
 تاريخ النشر 2021
  مجال البحث الهندسة المعلوماتية
والبحث باللغة English




اسأل ChatGPT حول البحث

We study black-box reward poisoning attacks against reinforcement learning (RL), in which an adversary aims to manipulate the rewards to mislead a sequence of RL agents with unknown algorithms to learn a nefarious policy in an environment unknown to the adversary a priori. That is, our attack makes minimum assumptions on the prior knowledge of the adversary: it has no initial knowledge of the environment or the learner, and neither does it observe the learners internal mechanism except for its performed actions. We design a novel black-box attack, U2, that can provably achieve a near-matching performance to the state-of-the-art white-box attack, demonstrating the feasibility of reward poisoning even in the most challenging black-box setting.



قيم البحث

اقرأ أيضاً

In reward-poisoning attacks against reinforcement learning (RL), an attacker can perturb the environment reward $r_t$ into $r_t+delta_t$ at each step, with the goal of forcing the RL agent to learn a nefarious policy. We categorize such attacks by th e infinity-norm constraint on $delta_t$: We provide a lower threshold below which reward-poisoning attack is infeasible and RL is certified to be safe; we provide a corresponding upper threshold above which the attack is feasible. Feasible attacks can be further categorized as non-adaptive where $delta_t$ depends only on $(s_t,a_t, s_{t+1})$, or adaptive where $delta_t$ depends further on the RL agents learning process at time $t$. Non-adaptive attacks have been the focus of prior works. However, we show that under mild conditions, adaptive attacks can achieve the nefarious policy in steps polynomial in state-space size $|S|$, whereas non-adaptive attacks require exponential steps. We provide a constructive proof that a Fast Adaptive Attack strategy achieves the polynomial rate. Finally, we show that empirically an attacker can find effective reward-poisoning attacks using state-of-the-art deep RL techniques.
We study a security threat to reinforcement learning where an attacker poisons the learning environment to force the agent into executing a target policy chosen by the attacker. As a victim, we consider RL agents whose objective is to find a policy t hat maximizes reward in infinite-horizon problem settings. The attacker can manipulate the rewards and the transition dynamics in the learning environment at training-time, and is interested in doing so in a stealthy manner. We propose an optimization framework for finding an optimal stealthy attack for different measures of attack cost. We provide lower/upper bounds on the attack cost, and instantiate our attacks in two settings: (i) an offline setting where the agent is doing planning in the poisoned environment, and (ii) an online setting where the agent is learning a policy with poisoned feedback. Our results show that the attacker can easily succeed in teaching any target policy to the victim under mild conditions and highlight a significant security threat to reinforcement learning agents in practice.
Recent work has discovered that deep reinforcement learning (DRL) policies are vulnerable to adversarial examples. These attacks mislead the policy of DRL agents by perturbing the state of the environment observed by agents. They are feasible in prin ciple but too slow to fool DRL policies in real time. We propose a new attack to fool DRL policies that is both effective and efficient enough to be mounted in real time. We utilize the Universal Adversarial Perturbation (UAP) method to compute effective perturbations independent of the individual inputs to which they are applied. Via an extensive evaluation using Atari 2600 games, we show that our technique is effective, as it fully degrades the performance of both deterministic and stochastic policies (up to 100%, even when the $l_infty$ bound on the perturbation is as small as 0.005). We also show that our attack is efficient, incurring an online computational cost of 0.027ms on average. It is faster compared to the response time (0.6ms on average) of agents with different DRL policies, and considerably faster than prior attacks (2.7ms on average). Furthermore, we demonstrate that known defenses are ineffective against universal perturbations. We propose an effective detection technique which can form the basis for robust defenses against attacks based on universal perturbations.
As machine learning systems grow in scale, so do their training data requirements, forcing practitioners to automate and outsource the curation of training data in order to achieve state-of-the-art performance. The absence of trustworthy human superv ision over the data collection process exposes organizations to security vulnerabilities; training data can be manipulated to control and degrade the downstream behaviors of learned models. The goal of this work is to systematically categorize and discuss a wide range of dataset vulnerabilities and exploits, approaches for defending against these threats, and an array of open problems in this space. In addition to describing various poisoning and backdoor threat models and the relationships among them, we develop their unified taxonomy.
In a poisoning attack, an adversary with control over a small fraction of the training data attempts to select that data in a way that induces a corrupted model that misbehaves in favor of the adversary. We consider poisoning attacks against convex m achine learning models and propose an efficient poisoning attack designed to induce a specified model. Unlike previous model-targeted poisoning attacks, our attack comes with provable convergence to {it any} attainable target classifier. The distance from the induced classifier to the target classifier is inversely proportional to the square root of the number of poisoning points. We also provide a lower bound on the minimum number of poisoning points needed to achieve a given target classifier. Our method uses online convex optimization, so finds poisoning points incrementally. This provides more flexibility than previous attacks which require a priori assumption about the number of poisoning points. Our attack is the first model-targeted poisoning attack that provides provable convergence for convex models, and in our experiments, it either exceeds or matches state-of-the-art attacks in terms of attack success rate and distance to the target model.

الأسئلة المقترحة

التعليقات
جاري جلب التعليقات جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
mircosoft-partner

هل ترغب بارسال اشعارات عن اخر التحديثات في شمرا-اكاديميا