In applying reinforcement learning (RL) to high-stakes domains, quantitative and qualitative evaluation using observational data can help practitioners understand the generalization performance of new policies. However, this type of off-policy evaluation (OPE) is inherently limited since offline data may not reflect the distribution shifts resulting from the application of new policies. On the other hand, online evaluation by collecting rollouts according to the new policy is often infeasible, as deploying new policies in these domains can be unsafe. In this work, we propose a semi-offline evaluation framework as an intermediate step between offline and online evaluation, where human users provide annotations of unobserved counterfactual trajectories. While tempting to simply augment existing data with such annotations, we show that this naive approach can lead to biased results. Instead, we design a new family of OPE estimators based on importance sampling (IS) and a novel weighting scheme that incorporate counterfactual annotations without introducing additional bias. We analyze the theoretical properties of our approach, showing its potential to reduce both bias and variance compared to standard IS estimators. Our analyses reveal important practical considerations for handling biased, noisy, or missing annotations. In a series of proof-of-concept experiments involving bandits and a healthcare-inspired simulator, we demonstrate that our approach outperforms purely offline IS estimators and is robust to imperfect annotations. Our framework, combined with principled human-centered design of annotation solicitation, can enable the application of RL in high-stakes domains.
相關內容
Federated learning (FL) has been widely studied recently due to its property to collaboratively train data from different devices without sharing the raw data. Nevertheless, recent studies show that an adversary can still be possible to infer private information about devices' data, e.g., sensitive attributes such as income, race, and sexual orientation. To mitigate the attribute inference attacks, various existing privacy-preserving FL methods can be adopted/adapted. However, all these existing methods have key limitations: they need to know the FL task in advance, or have intolerable computational overheads or utility losses, or do not have provable privacy guarantees. We address these issues and design a task-agnostic privacy-preserving presentation learning method for FL ({\bf TAPPFL}) against attribute inference attacks. TAPPFL is formulated via information theory. Specifically, TAPPFL has two mutual information goals, where one goal learns task-agnostic data representations that contain the least information about the private attribute in each device's data, and the other goal ensures the learnt data representations include as much information as possible about the device data to maintain FL utility. We also derive privacy guarantees of TAPPFL against worst-case attribute inference attacks, as well as the inherent tradeoff between utility preservation and privacy protection. Extensive results on multiple datasets and applications validate the effectiveness of TAPPFL to protect data privacy, maintain the FL utility, and be efficient as well. Experimental results also show that TAPPFL outperforms the existing defenses\footnote{Source code and full version: \url{//github.com/TAPPFL}}.
In multi-agent reinforcement learning (MARL), self-interested agents attempt to establish equilibrium and achieve coordination depending on game structure. However, existing MARL approaches are mostly bound by the simultaneous actions of all agents in the Markov game (MG) framework, and few works consider the formation of equilibrium strategies via asynchronous action coordination. In view of the advantages of Stackelberg equilibrium (SE) over Nash equilibrium, we construct a spatio-temporal sequential decision-making structure derived from the MG and propose an N-level policy model based on a conditional hypernetwork shared by all agents. This approach allows for asymmetric training with symmetric execution, with each agent responding optimally conditioned on the decisions made by superior agents. Agents can learn heterogeneous SE policies while still maintaining parameter sharing, which leads to reduced cost for learning and storage and enhanced scalability as the number of agents increases. Experiments demonstrate that our method effectively converges to the SE policies in repeated matrix game scenarios, and performs admirably in immensely complex settings including cooperative tasks and mixed tasks.
Decentralised learning has recently gained traction as an alternative to federated learning in which both data and coordination are distributed over its users. To preserve the confidentiality of users' data, decentralised learning relies on differential privacy, multi-party computation, or a combination thereof. However, running multiple privacy-preserving summations in sequence may allow adversaries to perform reconstruction attacks. Unfortunately, current reconstruction countermeasures either cannot trivially be adapted to the distributed setting, or add excessive amounts of noise. In this work, we first show that passive honest-but-curious adversaries can reconstruct other users' private data after several privacy-preserving summations. For example, in subgraphs with 18 users, we show that only three passive honest-but-curious adversaries succeed at reconstructing private data 11.0% of the time, requiring an average of 8.8 summations per adversary. The success rate is independent of the size of the full network. We consider weak adversaries, who do not control the graph topology and can exploit neither the workings of the summation protocol nor the specifics of users' data. We develop a mathematical understanding of how reconstruction relates to topology and propose the first topology-based decentralised defence against reconstruction attacks. Specifically, we show that reconstruction requires a number of adversaries linear in the length of the network's shortest cycle. Consequently, reconstructing private data from privacy-preserving summations is impossible in acyclic networks. Our work is a stepping stone for a formal theory of decentralised reconstruction defences based on topology. Such a theory would generalise our countermeasure beyond summation, define confidentiality in terms of entropy, and describe the effects of (topology-aware) differential privacy.
Deep reinforcement learning (RL) provides powerful methods for training optimal sequential decision-making agents. As collecting real-world interactions can entail additional costs and safety risks, the common paradigm of sim2real conducts training in a simulator, followed by real-world deployment. Unfortunately, RL agents easily overfit to the choice of simulated training environments, and worse still, learning ends when the agent masters the specific set of simulated environments. In contrast, the real world is highly open-ended, featuring endlessly evolving environments and challenges, making such RL approaches unsuitable. Simply randomizing over simulated environments is insufficient, as it requires making arbitrary distributional assumptions and can be combinatorially less likely to sample specific environment instances that are useful for learning. An ideal learning process should automatically adapt the training environment to maximize the learning potential of the agent over an open-ended task space that matches or surpasses the complexity of the real world. This thesis develops a class of methods called Unsupervised Environment Design (UED), which aim to produce such open-ended processes. Given an environment design space, UED automatically generates an infinite sequence or curriculum of training environments at the frontier of the learning agent's capabilities. Through extensive empirical studies and theoretical arguments founded on minimax-regret decision theory and game theory, the findings in this thesis show that UED autocurricula can produce RL agents exhibiting significantly improved robustness and generalization to previously unseen environment instances. Such autocurricula are promising paths toward open-ended learning systems that achieve more general intelligence by continually generating and mastering additional challenges of their own design.
Inductive Conformal Prediction (ICP) provides a practical and effective approach for equipping deep learning models with uncertainty estimates in the form of set-valued predictions which are guaranteed to contain the ground truth with high probability. Despite the appeal of this coverage guarantee, these sets may not be efficient: the size and contents of the prediction sets are not directly controlled, and instead depend on the underlying model and choice of score function. To remedy this, recent work has proposed learning model and score function parameters using data to directly optimize the efficiency of the ICP prediction sets. While appealing, the generalization theory for such an approach is lacking: direct optimization of empirical efficiency may yield prediction sets that are either no longer efficient on test data, or no longer obtain the required coverage on test data. In this work, we use PAC-Bayes theory to obtain generalization bounds on both the coverage and the efficiency of set-valued predictors which can be directly optimized to maximize efficiency while satisfying a desired test coverage. In contrast to prior work, our framework allows us to utilize the entire calibration dataset to learn the parameters of the model and score function, instead of requiring a separate hold-out set for obtaining test-time coverage guarantees. We leverage these theoretical results to provide a practical algorithm for using calibration data to simultaneously fine-tune the parameters of a model and score function while guaranteeing test-time coverage and efficiency of the resulting prediction sets. We evaluate the approach on regression and classification tasks, and outperform baselines calibrated using a Hoeffding bound-based PAC guarantee on ICP, especially in the low-data regime.
Recent contrastive representation learning methods rely on estimating mutual information (MI) between multiple views of an underlying context. E.g., we can derive multiple views of a given image by applying data augmentation, or we can split a sequence into views comprising the past and future of some step in the sequence. Contrastive lower bounds on MI are easy to optimize, but have a strong underestimation bias when estimating large amounts of MI. We propose decomposing the full MI estimation problem into a sum of smaller estimation problems by splitting one of the views into progressively more informed subviews and by applying the chain rule on MI between the decomposed views. This expression contains a sum of unconditional and conditional MI terms, each measuring modest chunks of the total MI, which facilitates approximation via contrastive bounds. To maximize the sum, we formulate a contrastive lower bound on the conditional MI which can be approximated efficiently. We refer to our general approach as Decomposed Estimation of Mutual Information (DEMI). We show that DEMI can capture a larger amount of MI than standard non-decomposed contrastive bounds in a synthetic setting, and learns better representations in a vision domain and for dialogue generation.
In semi-supervised domain adaptation, a few labeled samples per class in the target domain guide features of the remaining target samples to aggregate around them. However, the trained model cannot produce a highly discriminative feature representation for the target domain because the training data is dominated by labeled samples from the source domain. This could lead to disconnection between the labeled and unlabeled target samples as well as misalignment between unlabeled target samples and the source domain. In this paper, we propose a novel approach called Cross-domain Adaptive Clustering to address this problem. To achieve both inter-domain and intra-domain adaptation, we first introduce an adversarial adaptive clustering loss to group features of unlabeled target data into clusters and perform cluster-wise feature alignment across the source and target domains. We further apply pseudo labeling to unlabeled samples in the target domain and retain pseudo-labels with high confidence. Pseudo labeling expands the number of ``labeled" samples in each class in the target domain, and thus produces a more robust and powerful cluster core for each class to facilitate adversarial learning. Extensive experiments on benchmark datasets, including DomainNet, Office-Home and Office, demonstrate that our proposed approach achieves the state-of-the-art performance in semi-supervised domain adaptation.
This paper aims to mitigate straggler effects in synchronous distributed learning for multi-agent reinforcement learning (MARL) problems. Stragglers arise frequently in a distributed learning system, due to the existence of various system disturbances such as slow-downs or failures of compute nodes and communication bottlenecks. To resolve this issue, we propose a coded distributed learning framework, which speeds up the training of MARL algorithms in the presence of stragglers, while maintaining the same accuracy as the centralized approach. As an illustration, a coded distributed version of the multi-agent deep deterministic policy gradient(MADDPG) algorithm is developed and evaluated. Different coding schemes, including maximum distance separable (MDS)code, random sparse code, replication-based code, and regular low density parity check (LDPC) code are also investigated. Simulations in several multi-robot problems demonstrate the promising performance of the proposed framework.
We advocate the use of implicit fields for learning generative models of shapes and introduce an implicit field decoder for shape generation, aimed at improving the visual quality of the generated shapes. An implicit field assigns a value to each point in 3D space, so that a shape can be extracted as an iso-surface. Our implicit field decoder is trained to perform this assignment by means of a binary classifier. Specifically, it takes a point coordinate, along with a feature vector encoding a shape, and outputs a value which indicates whether the point is outside the shape or not. By replacing conventional decoders by our decoder for representation learning and generative modeling of shapes, we demonstrate superior results for tasks such as shape autoencoding, generation, interpolation, and single-view 3D reconstruction, particularly in terms of visual quality.
Deep learning has yielded state-of-the-art performance on many natural language processing tasks including named entity recognition (NER). However, this typically requires large amounts of labeled data. In this work, we demonstrate that the amount of labeled training data can be drastically reduced when deep learning is combined with active learning. While active learning is sample-efficient, it can be computationally expensive since it requires iterative retraining. To speed this up, we introduce a lightweight architecture for NER, viz., the CNN-CNN-LSTM model consisting of convolutional character and word encoders and a long short term memory (LSTM) tag decoder. The model achieves nearly state-of-the-art performance on standard datasets for the task while being computationally much more efficient than best performing models. We carry out incremental active learning, during the training process, and are able to nearly match state-of-the-art performance with just 25\% of the original training data.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191