United States

Adaptive optimization algorithms such as Adam (Kingma and Ba, 2014) are widely used in deep learning. The stability of such algorithms is often improved with a warmup schedule for the learning rate. Motivated by the difficulty of choosing and tuning warmup schedules, recent work proposes automatic variance rectification of Adam&#39;s adaptive learning rate, claiming that this rectified approach (&quot;RAdam&quot;) surpasses the vanilla Adam algorithm and reduces the need for expensive tuning of Adam with warmup. In this work, we refute this analysis and provide an alternative explanation for the necessity of warmup based on the magnitude of the update term, which is of greater relevance to training stability. We then provide some &quot;rule-of-thumb&quot; warmup schedules, and we demonstrate that simple untuned warmup of Adam performs more-or-less identically to RAdam in typical practical settings. We conclude by suggesting that practitioners stick to linear warmup with Adam, with a sensible default being linear warmup over 2 / (1 - Î²â‚‚) training iterations.

AAAI 2021

On the Adequacy of Untuned Warmup for Adaptive Optimization

The purpose of the AAAI conference is to promote research in artificial intelligence (AI) and scientific exchange among AI researchers, practitioners, scientists, and engineers in affiliated disciplines. 

Adaptive optimization algorithms such as Adam (Kingma and Ba, 2014) are widely used in deep learning. The stability of such algorithms is often improved with a warmup schedule for the learning rate. Motivated by the difficulty of choosing and tuning warmup schedules, recent work proposes automatic variance rectification of Adam's adaptive learning rate, claiming that this rectified approach ("RAdam") surpasses the vanilla Adam algorithm and reduces the need for expensive tuning of Adam with warmup. In this work, we refute this analysis and provide an alternative explanation for the necessity of warmup based on the magnitude of the update term, which is of greater relevance to training stability. We then provide some "rule-of-thumb" warmup schedules, and we demonstrate that simple untuned warmup of Adam performs more-or-less identically to RAdam in typical practical settings. We conclude by suggesting that practitioners stick to linear warmup with Adam, with a sensible default being linear warmup over 2 / (1 - Î²â‚‚) training iterations.

technical paper

We focus on resilience in cooperative multi-agent systems, where agents can change their behavior due to udpates or failures of hardware and software components. Current state-of-the-art approaches to cooperative multi-agent reinforcement learning (MARL) have either focused on idealized settings without any changes or on very specialized scenarios, where the number of changing agents is fixed, e.g., in extreme cases with only one productive agent. Therefore, we propose Resilient Adversarial value Decomposition with Antagonist-Ratios (RADAR). RADAR offers a value decomposition scheme to train competing teams of varying size for improved resilience against arbitrary agent changes. We evaluate RADAR in two cooperative multi-agent domains and show that RADAR achieves better worst case performance w.r.t. arbitrary agent changes than state-of-the-art MARL.

Resilient Multi-Agent Reinforcement Learning with Adversarial Value Decomposition

Convolutional neural networks (CNNs) have achieved state-of-the-art performance on various tasks in computer vision. However, recent studies demonstrate that these models are vulnerable to carefully crafted adversarial samples and suffer from a significant performance drop when predicting them. Many methods have been proposed to improve adversarial robustness (e.g., adversarial training and new loss functions to learn adversarially robust feature representations). Here we offer a unique insight into the predictive behavior of CNNs that they tend to misclassify adversarial samples into the most probable false classes. This inspires us to propose a new Probabilistically Compact (PC) loss with logit constraints which can be used as a drop-in replacement for cross-entropy (CE) loss to improve CNN's adversarial robustness. Specifically, PC loss enlarges the probability gaps between true class and false classes meanwhile the logit constraints prevent the gaps from being melted by a small perturbation. We extensively compare our method with the state-of-the-art using large scale datasets under both white-box and black-box attacks to demonstrate its effectiveness. The source codes are available at https://github.com/xinli0928/PC-LC.

Improving Adversarial Robustness via Probabilistically Compact Loss with Logit Constraints

Automated story plot generation is the task of generating a coherent sequence of plot events. Causal relations between plot events are believed to increase the perception of story and plot coherence. In this work, we introduce the concept of soft causal relations as causal relations inferred from commonsense reasoning. We demonstrate C2PO, an approach to narrative generation that operationalizes this concept through Causal, Commonsense Plot Ordering. Using human-participant protocols, we evaluate our system against baseline systems with different commonsense reasoning reasoning and inductive biases to determine the role of soft causal relations in perceived story quality. Through these studies we also probe the interplay of how changes in commonsense norms across storytelling genres affect perceptions of story quality.

Automated Storytelling via Causal, Commonsense Plot Ordering

Training sophisticated machine learning models usually requires many training samples. Especially in healthcare settings these samples can be very expensive, meaning that one institution alone usually does not have enough. Merging privacy-sensitive data from different sources is usually restricted by data security and data protection measures. This can lead to approaches that reduce data quality by putting noise onto the variables (e.g., in epsilon-differential privacy) or omitting certain values (e.g., for k-anonymity). Other measures based on cryptographic methods can lead to very time-consuming computations, which is especially problematic for larger multi-omics data. We address this problem by introducing ESCAPED, which stands for Efficient SeCure And PrivatE Dot product framework. ESCAPED enables the computation of the dot product of vectors from multiple sources on a third-party, which later trains kernel-based machine learning algorithms, while neither sacrificing privacy nor adding noise. We have evaluated our framework on drug resistance prediction for HIV-infected people and multi-omics dimensionality reduction and clustering problems in precision medicine. In terms of execution time, our framework significantly outperforms the best-fitting existing approaches without sacrificing the performance of the algorithm. Even though we only present the benefit for kernel-based algorithms, our framework can open up new research opportunities for further machine learning models that require the dot product of vectors from multiple sources.

ESCAPED: Efficient Secure and Private Dot Product Framework for Kernel-Based Machine Learning Algorithms with Applications in Healthcare

We propose a deep generative framework for multi-view learning
based on a probabilistic interpretation of canonical correlation
analysis (CCA). The model combines a linear multi-view
layer in the latent space with deep generative networks as
observation models, to decompose the variability in multiple
views into a shared latent representation that describes the
common underlying sources of variation and a set of viewspecific
components. To approximate the posterior distribution
of the latent multi-view layer, an efficient variational inference
procedure is developed based on the solution of probabilistic
CCA. The model is then generalized to an arbitrary number of
views. An empirical analysis confirms that the proposed deep
multi-view model can discover subtle relationships between
multiple views and recover rich representations.

Deep Probabilistic Canonical Correlation Analysis

The teacher-student framework aims to improve the sample efficiency of RL algorithms by deploying an advising mechanism in which a teacher helps a student by guiding its exploration. Prior work in this field has considered an advising mechanism where the teacher advises the student about the optimal action to take in a given state. However, real-world teachers can leverage domain expertise to provide more informative signals. Using this insight, we propose to extend the current advising framework wherein the teacher would provide not only the optimal action but also a qualitative assessment of the state. We introduce a novel architecture, namely Advice Replay Memory (ARM), to effectively reuse the advice provided by the teacher. We demonstrate the robustness of our approach by showcasing our experiments on multiple Atari 2600 games using a fixed set of hyper-parameters. Additionally, we show that a student taking help even from a sub-optimal teacher can achieve significant performance boosts and eventually outperform the teacher. Our approach outperforms the baselines even when provided with comparatively suboptimal teachers and an advising budget, which is smaller by orders of magnitude. The contributions of our paper are 4-fold (a) effectively leveraging a teacher's knowledge by richer advising (b) introduction of ARM to effectively reuse the advice throughout learning (c) ability to achieve significant performance boost even with a coarse state categorization (d) enabling the student to outperform the teacher.

An Enhanced Advising Model in Teacher-Student Framework Using State Categorization

We study the problem of automated mechanism design with partial verification, where each type can (mis)report only a restricted set of types (rather than any other type), induced by the principal's limited verification power. We prove hardness results when the revelation principle does not necessarily hold, as well as when types have even minimally different preferences. In light of these hardness results, we focus on truthful mechanisms in the setting where all types share the same preference over outcomes, which is motivated by applications in, e.g., strategic classification. We present a number of algorithmic and structural results, including an efficient algorithm for finding optimal deterministic truthful mechanisms, which also implies a faster algorithm for finding optimal randomized truthful mechanisms via a characterization based on the notion of convexity. We then consider a more general setting, where the principal's cost is a function of the combination of outcomes assigned to each type. In particular, we focus on the case where the cost function is submodular, and give generalizations of essentially all our results in the classical setting where the cost function is additive. Our results provide a relatively complete picture for automated mechanisms design with partial verification.

Automated Mechanism Design for Classification with Partial Verification

Attribution methods have been developed to understand the decision making process of machine learning models, especially deep neural networks, by assigning importance scores to individual features. Existing attribution methods often built upon empirical intuitions and heuristics. There still lacks a general and theoretical framework that not only can unify these attribution methods, but also theoretically reveal their rationales, fidelity, and limitations. To bridge the gap, in this paper, we propose a Taylor attribution framework and reformulate seven mainstream attribution methods into the framework. Based on reformulations, we analyze the attribution methods in terms of rationale, fidelity, and limitation.
Moreover, We establish three principles for a good attribution in the Taylor attribution framework, i.e., low approximation error, correct contribution assignment, and unbiased baseline selection. Finally, we empirically validate the Taylor reformulations, and reveal a positive correlation between the attribution performance and the number of principles followed by the attribution method via benchmarking on real-world datasets.

A Unified Taylor Framework for Revisiting Attribution Methods

This paper proposes a new paradigm for learning a set of independent logical rules in disjunctive normal form as an interpretable model for classification. We consider the problem of learning an interpretable decision rule set as training a neural network in a specific, yet very simple two-layer architecture. Each neuron in the first layer directly maps to an interpretable if-then rule after training, and the output neuron in the second layer directly maps to a disjunction of the first layer rules to form the decision rule set. Our representation of neurons in this first rules layer enables us to encode both the positive and the negative association of features in a decision rule. State-of-the-art neural net training approaches can be leveraged for learning highly accurate classification models. Moreover, we propose a sparsity-based regularization approach to balance between classification accuracy and the simplicity of the derived rules. Our experimental results show that our method can generate more accurate decision rule sets than other state-of-the-art rule-learning algorithms with better accuracy-simplicity trade-offs. Further, when compared with uninterpretable black-box machine learning approaches such as random forests and full-precision deep neural networks, our approach can easily find interpretable decision rule sets that have comparable predictive performance.

Learning Accurate and Interpretable Decision Rule Sets from Neural Networks

Reinforcement learning in complex environments may require supervision to prevent the agent from attempting dangerous actions. As a result of supervisor intervention, the executed action may differ from the action specified by the policy. How does this affect learning? We present the Modified-Action Markov Decision Process, an extension of the MDP model that allows actions to differ from the policy. We analyze the asymptotic behaviours of common reinforcement learning algorithms in this setting and show that they adapt in different ways: some completely ignore modifications while others go to various lengths in trying to avoid action modifications that decrease reward. By choosing the right algorithm, developers can prevent their agents from learning to circumvent interruptions or constraints, and better control agent responses to other kinds of action modification, like self-damage.

Downloads

Next from AAAI 2021

Resilient Multi-Agent Reinforcement Learning with Adversarial Value Decomposition

Similar lecture

Large Batch Optimization for Deep Learning Using New Complete Layer-Wise Adaptive Rate Scaling

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAAI 2021

Resilient Multi-Agent Reinforcement Learning with Adversarial Value Decomposition

Similar lecture

Large Batch Optimization for Deep Learning Using New Complete Layer-Wise Adaptive Rate Scaling

Downloads