ICML 2022 Papers

What is the best way to exploit extra data – be it unlabeled data from the same task, or labeled data from a related task – to learn a given task? This paper formalizes the question using the theory of reference priors. Reference priors are objective, uninformative Bayesian priors that maximize the mutual information between the task and the weights of the model. Such priors enable the task to maximally affect the Bayesian posterior, e.g., reference priors depend upon the number of samples available for learning the task and for very small sample sizes, the prior puts more probability mass on low-complexity models in the hypothesis space. This paper presents the first demonstration of reference priors for medium-scale deep networks and image-based data. We develop generalizations of reference priors and demonstrate applications to two problems. First, by using unlabeled data to compute the reference prior, we develop new Bayesian semi-supervised learning methods that remain effective even with very few samples per class. Second, by using labeled data from the source task to compute the reference prior, we develop a new pretraining method for transfer learning that allows data from the target task to maximally affect the Bayesian posterior. Empirical validation of these methods is conducted on image classification datasets. Code is available at https://github.com/grasp-lyrl/deep_reference_priors

Reliable probability estimation is of crucial importance in many real-world applications where there is inherent (aleatoric) uncertainty. Probability-estimation models are trained on observed outcomes (e.g. whether it has rained or not, or whether a patient has died or not), because the ground-truth probabilities of the events of interest are typically unknown. The problem is therefore analogous to binary classification, with the difference that the objective is to estimate probabilities rather than predicting the specific outcome. This work investigates probability estimation from high-dimensional data using deep neural networks. There exist several methods to improve the probabilities generated by these models but they mostly focus on model (epistemic) uncertainty. For problems with inherent uncertainty, it is challenging to evaluate performance without access to ground-truth probabilities. To address this, we build a synthetic dataset to study and compare different computable metrics. We evaluate existing methods on the synthetic data as well as on three real-world probability estimation tasks, all of which involve inherent uncertainty: precipitation forecasting from radar images, predicting cancer patient survival from histopathology images, and predicting car crashes from dashcam videos. We also give a theoretical analysis of a model for high-dimensional probability estimation which reproduces several of the phenomena evinced in our experiments. Finally, we propose a new method for probability estimation using neural networks, which modifies the training process to promote output probabilities that are consistent with empirical probabilities computed from the data. The method outperforms existing approaches on most metrics on the simulated as well as real-world data.

Model fusion without accessing training data in machine learning has attracted increasing interest due to the practical resource-saving and data privacy issues. During the training process, the neural weights of each model can be randomly permuted, and we have to align the channels of each layer before fusing them. Regrading the channels as nodes and weights as edges, aligning the channels to maximize weight similarity is a challenging NP-hard assignment problem. Due to its quadratic assignment nature, we formulate the model fusion problem as a graph matching task, considering the second-order similarity of model weights instead of previous work merely formulating model fusion as a linear assignment problem. For the rising problem scale and multi-model consistency issues, we propose an efficient graduated assignment-based model fusion method, dubbed GAMF, which iteratively updates the matchings in a consistency-maintaining manner. We apply GAMF to tackle the compact model ensemble task and federated learning task on MNIST, CIFAR-10, CIFAR-100, and Tiny-Imagenet. The performance shows the efficacy of our GAMF compared to state-of-the-art baselines.

We systematize the approach to the investigation of deep neural network landscapes by basing it on the geometry of the space of implemented functions rather than the space of parameters. Grouping classifiers into equivalence classes, we develop a standardized parameterization in which all symmetries are removed, resulting in a toroidal topology. On this space, we explore the error landscape rather than the loss. This lets us derive a meaningful notion of the flatness of minimizers and of the geodesic paths connecting them. Using different optimization algorithms that sample minimizers with different flatness we study the mode connectivity and relative distances. Testing a variety of state-of-the-art architectures and benchmark datasets, we confirm the correlation between flatness and generalization performance; we further show that in function space flatter minima are closer to each other and that the barriers along the geodesics connecting them are small. We also find that minimizers found by variants of gradient descent can be connected by zero-error paths composed of two straight lines in parameter space, i.e. polygonal chains with a single bend. We observe similar qualitative results in neural networks with binary weights and activations, providing one of the first results concerning the connectivity in this setting. Our results hinge on symmetry removal, and are in remarkable agreement with the rich phenomenology described by some recent analytical studies performed on simple shallow models.

One of the arguments to explain the success of deep learning is the powerful approximation capacity of deep neural networks. Such capacity is generally accompanied by the explosive growth of the number of parameters, which, in turn, leads to high computational costs. It is of great interest to ask whether we can achieve successful deep learning with a small number of learnable parameters adapting to the target function. From an approximation perspective, this paper shows that the number of parameters that need to be learned can be significantly smaller than people typically expect. First, we theoretically design ReLU networks with a few learnable parameters to achieve an attractive approximation. We prove by construction that, for any Lipschitz continuous function $f$ on $[0,1]^d$ with a Lipschitz constant $\lambda>0$, a ReLU network with $n+2$ intrinsic parameters (those depending on $f$) can approximate $f$ with an exponentially small error $5 \lambda \sqrt{d} \, 2^{-n}$. Such a result is generalized to generic continuous functions. Furthermore, we show that the idea of learning a small number of parameters to achieve a good approximation can be numerically observed. We conduct several experiments to verify that training a small part of parameters can also achieve good results for classification problems if other parameters are pre-specified or pre-trained from a related problem.

Mean rewards of actions are often correlated. The form of these correlations may be complex and unknown a priori, such as the preferences of users for recommended products and their categories. To maximize statistical efficiency, it is important to leverage these correlations when learning. We formulate a bandit variant of this problem where the correlations of mean action rewards are represented by a hierarchical Bayesian model with latent variables. Since the hierarchy can have multiple layers, we call it deep. We propose a hierarchical Thompson sampling algorithm (HierTS) for this problem and show how to implement it efficiently for Gaussian hierarchies. The efficient implementation is possible due to a novel exact hierarchical representation of the posterior, which itself is of independent interest. We use this exact posterior to analyze the Bayes regret of HierTS. Our regret bounds reflect the structure of the problem, that the regret decreases with more informative priors, and can be recast to highlight reduced dependence on the number of actions. We confirm these theoretical findings empirically, in both synthetic and real-world experiments.

Deep metric learning aims to learn distance metrics that measure similarities and dissimilarities between samples. The existing approaches typically focus on designing different hard sample mining or distance margin strategies and then minimize a pair/triplet-based or proxy-based loss over the training data. However, this can lead the model to recklessly learn all the correlated distances found in training data including the spurious distance (e.g., background differences) that is not the distance of interest and can harm the generalization of the learned metric. To address this issue, we study metric learning from a causality perspective and accordingly propose deep causal metric learning (DCML) that pursues the true causality of the distance between samples. DCML is achieved through explicitly learning environment-invariant attention and task-invariant embedding based on causal inference. Extensive experiments on several benchmark datasets demonstrate the superiority of DCML over the existing methods.

Past work has shown that large language models are susceptible to privacy attacks, where adversaries generate sequences from a trained model and detect which sequences are memorized from the training set. In this work, we show that the success of these attacks is largely due to duplication in commonly used web-scraped training sets. We first show that the rate at which language models regenerate training sequences is superlinearly related to a sequence’s count in the training set. For instance, a sequence that is present 10 times in the training data is on average generated 1000x more often than a sequence that is present only once. We next show that existing methods for detecting memorized sequences have near-chance accuracy on non-duplicated training sequences. Finally, we find that after applying methods to deduplicate training data, language models are considerably more secure against these types of privacy attacks. Taken together, our results motivate an increased focus on deduplication in privacy-sensitive applications and a reevaluation of the practicality of existing privacy attacks.

Value decomposition (VD) methods have been widely used in cooperative multi-agent reinforcement learning (MARL), where credit assignment plays an important role in guiding the agents’ decentralized execution. In this paper, we investigate VD from a novel perspective of causal inference. We first show that the environment in existing VD methods is an unobserved confounder as the common cause factor of the global state and the joint value function, which leads to the confounding bias on learning credit assignment. We then present our approach, deconfounded value decomposition (DVD), which cuts off the backdoor confounding path from the global state to the joint value function. The cut is implemented by introducing the

High-order interactions between multiple objects are common in real-world applications. Although tensor decomposition is a popular framework for high-order interaction analysis and prediction, most methods cannot well exploit the valuable timestamp information in data. The existent methods either discard the timestamps or convert them into discrete steps or use over-simplistic decomposition models. As a result, these methods might not be capable enough of capturing complex, fine-grained temporal dynamics or making accurate predictions for long-term interaction results. To overcome these limitations, we propose a novel Temporal High-order Interaction decompoSition model based on Ordinary Differential Equations (THIS-ODE). We model the time-varying interaction result with a latent ODE. To capture the complex temporal dynamics, we use a neural network (NN) to learn the time derivative of the ODE state. We use the representation of the interaction objects to model the initial value of the ODE and to constitute a part of the NN input to compute the state. In this way, the temporal relationships of the participant objects can be estimated and encoded into their representations. For tractable and scalable inference, we use forward sensitivity analysis to efficiently compute the gradient of ODE state, based on which we use integral transform to develop a stochastic mini-batch learning algorithm. We demonstrate the advantage of our approach in simulation and four real-world applications.

In the last years decision-focused learning framework, also known as predict-and-optimize, have received increasing attention. In this setting, the predictions of a machine learning model are used as estimated cost coefficients in the objective function of a discrete combinatorial optimization problem for decision making. Decision-focused learning proposes to train the ML models, often neural network models, by directly optimizing the quality of decisions made by the optimization solvers. Based on a recent work that proposed a noise contrastive estimation loss over a subset of the solution space, we observe that decision-focused learning can more generally be seen as a learning-to-rank problem, where the goal is to learn an objective function that ranks the feasible points correctly. This observation is independent of the optimization method used and of the form of the objective function. We develop pointwise, pairwise and listwise ranking loss functions, which can be differentiated in closed form given a subset of solutions. We empirically investigate the quality of our generic methods compared to existing decision-focused learning approaches with competitive results. Furthermore, controlling the subset of solutions allows controlling the runtime considerably, with limited effect on regret.

This paper proposes a theoretical analysis of a Lasso-based classification algorithm. Leveraging on a realistic regime where the dimension of the data $p$ and their number $n$ are of the same order of magnitude, the theoretical classification error is derived as a function of the data statistics. As a result, insights into the functioning of the Lasso in classification and its differences with competing algorithms are highlighted. Our work is based on an original novel analysis of the Iterative Soft-Thresholding Algorithm (ISTA), which may be of independent interest beyond the particular problem studied here and may be adapted to similar iterative schemes. A theoretical optimization of the model’s hyperparameters is also provided, which allows for the data- and time-consuming cross-validation to be avoided. Finally, several applications on synthetic and real data are provided to validate the theoretical study and justify its impact in the design and understanding of algorithms of practical interest.

We study the problem of networked online convex optimization, where each agent individually decides on an action at every time step and agents cooperatively seek to minimize the total global cost over a finite horizon. The global cost is made up of three types of local costs: convex node costs, temporal interaction costs, and spatial interaction costs. In deciding their individual action at each time, an agent has access to predictions of local cost functions for the next $k$ time steps in an $r$-hop neighborhood. Our work proposes a novel online algorithm, Localized Predictive Control (LPC), which generalizes predictive control to multi-agent systems. We show that LPC achieves a competitive ratio of $1 + \tilde{O}(\rho_T^k) + \tilde{O}(\rho_S^r)$ in an adversarial setting, where $\rho_T$ and $\rho_S$ are constants in $(0, 1)$ that increase with the relative strength of temporal and spatial interaction costs, respectively. This is the first competitive ratio bound on decentralized predictive control for networked online convex optimization. Further, we show that the dependence on $k$ and $r$ in our results is near optimal by lower bounding the competitive ratio of any decentralized online algorithm.

Estimating optimal transport (OT) maps (a.k.a. Monge maps) between two measures P and Q is a problem fraught with computational and statistical challenges. A promising approach lies in using the dual potential functions obtained when solving an entropy-regularized OT problem between samples P_n and Q_n, which can be used to recover an approximately optimal map. The negentropy penalization in that scheme introduces, however, an estimation bias that grows with the regularization strength. A well-known remedy to debias such estimates, which has gained wide popularity among practitioners of regularized OT, is to center them, by subtracting auxiliary problems involving P_n and itself, as well as Q_n and itself. We do prove that, under favorable conditions on P and Q, debiasing can yield better approximations to the Monge map. However, and perhaps surprisingly, we present a few cases in which debiasing is provably detrimental in a statistical sense, notably when the regularization strength is large or the number of samples is small. These claims are validated experimentally on synthetic and real datasets, and should reopen the debate on whether debiasing is needed when using entropic OT.

Tandem mass spectrometry is the only high-throughput method for analyzing the protein content of complex biological samples and is thus the primary technology driving the growth of the field of proteomics. A key outstanding challenge in this field involves identifying the sequence of amino acids -the peptide- responsible for generating each observed spectrum, without making use of prior knowledge in the form of a peptide sequence database. Although various machine learning methods have been developed to address this de novo sequencing problem, challenges that arise when modeling tandem mass spectra have led to complex models that combine multiple neural networks and post-processing steps. We propose a simple yet powerful method for de novo peptide sequencing, Casanovo, that uses a transformer framework to map directly from a sequence of observed peaks (a mass spectrum) to a sequence of amino acids (a peptide). Our experiments show that Casanovo achieves state-of-the-art performance on a benchmark dataset using a standard cross-species evaluation framework which involves testing with spectra with never-before-seen peptide labels. Casanovo not only achieves superior performance but does so at a fraction of the model complexity and inference time required by other methods.

Recent studies have demonstrated that gradient matching-based dataset synthesis, or dataset condensation (DC), methods can achieve state-of-theart performance when applied to data-efficient learning tasks. However, in this study, we prove that the existing DC methods can perform worse than the random selection method when taskirrelevant information forms a significant part of the training dataset. We attribute this to the lack of participation of the contrastive signals between the classes resulting from the class-wise gradient matching strategy. To address this problem, we propose Dataset Condensation with Contrastive signals (DCC) by modifying the loss function to enable the DC methods to effectively capture the differences between classes. In addition, we analyze the new loss function in terms of training dynamics by tracking the kernel velocity. Furthermore, we introduce a bi-level warm-up strategy to stabilize the optimization. Our experimental results indicate that while the existing methods are ineffective for fine-grained image classification tasks, the proposed method can successfully generate informative synthetic datasets for the same tasks. Moreover, we demonstrate that the proposed method outperforms the baselines even on benchmark datasets such as SVHN, CIFAR-10, and CIFAR-100. Finally, we demonstrate the high applicability of the proposed method by applying it to continual learning tasks.

The great success of machine learning with massive amounts of data comes at a price of huge computation costs and storage for training and tuning. Recent studies on dataset condensation attempt to reduce the dependence on such massive data by synthesizing a compact training dataset. However, the existing approaches have fundamental limitations in optimization due to the limited representability of synthetic datasets without considering any data regularity characteristics. To this end, we propose a novel condensation framework that generates multiple synthetic data with a limited storage budget via efficient parameterization considering data regularity. We further analyze the shortcomings of the existing gradient matching-based condensation methods and develop an effective optimization technique for improving the condensation of training data information. We propose a unified algorithm that drastically improves the quality of condensed data against the current state-of-the-art on CIFAR-10, ImageNet, and Speech Commands.

We present a conceptual framework,

Systematic quantification of data quality is critical for consistent model performance. Prior works have focused on out-of-distribution data. Instead, we tackle an understudied yet equally important problem of characterizing incongruous regions of in-distribution (ID) data, which may arise from feature space heterogeneity. To this end, we propose a paradigm shift with Data-SUITE: a data-centric AI framework to identify these regions, independent of a task-specific model. Data-SUITE leverages copula modeling, representation learning, and conformal prediction to build feature-wise confidence interval estimators based on a set of training instances. These estimators can be used to evaluate the congruence of test instances with respect to the training set, to answer two practically useful questions: (1) which test instances will be reliably predicted by a model trained with the training instances? and (2) can we identify incongruous regions of the feature space so that data owners understand the data’s limitations or guide future data collection? We empirically validate Data-SUITE’s performance and coverage guarantees and demonstrate on cross-site medical data, biased data, and data with concept drift, that Data-SUITE best identifies ID regions where a downstream model may be reliable (independent of said model). We also illustrate how these identified regions can provide insights into datasets and highlight their limitations.