ICML 2024 Papers

Performative prediction, as introduced by Perdomo et al., is a framework for studying social prediction in which the data distribution itself changes in response to the deployment of a model. Existing work in this field usually hinges on three assumptions that are easily violated in practice: that the performative risk is convex over the deployed model, that the mapping from the model to the data distribution is known to the model designer in advance, and the first-order information of the performative risk is available. In this paper, we initiate the study of performative prediction problems that do not require these assumptions. Specifically, we develop a parameterization framework that parametrizes the performative prediction objective as a function of the induced data distribution. We also develop a two-level zeroth-order optimization procedure, where the first level performs iterative optimization on the distribution parameter space, and the second level learns the model that induced a particular target distribution parameter at each iteration. Under mild conditions, this reparameterization allows us to transform the non-convex objective into a convex one and achieve provable regret guarantees. In particular, we provide a regret bound that is sublinear in the total number of performative samples taken and is only polynomial in the dimension of the model parameter.

Optimization layers in deep neural networks have enjoyed a growing popularity in structured learning, improving the state of the art on a variety of applications. Yet, these pipelines lack interpretability since they are made of two opaque layers: a highly non-linear prediction model, such as a deep neural network, and an optimization layer, which is typically a complex black-box solver. Our goal is to improve the transparency of such methods by providing counterfactual explanations. We build upon variational autoencoders a principled way of obtaining counterfactuals: working in the latent space leads to a natural notion of plausibility of explanations. We finally introduce a variant of the classic loss for VAE training that improves their performance in our specific structured context. These provide the foundations of CF-OPT, a first-order optimization algorithm that can find counterfactual explanations for a broad class of structured learning architectures. Our numerical results show that both close and plausible explanations can be obtained for problems from the recent literature.

Conformal prediction (CP) has been a popular method for uncertainty quantification because it is distribution-free, model-agnostic, and theoretically sound. For forecasting problems in supervised learning, most CP methods focus on building prediction intervals for univariate responses. In this work, we develop a sequential CP method called $\texttt{MultiDimSPCI}$ that builds prediction $\textit{regions}$ for a multivariate response, especially in the context of multivariate time series, which are not exchangeable. Theoretically, we estimate $\textit{finite-sample}$ high-probability bounds on the conditional coverage gap. Empirically, we demonstrate that $\texttt{MultiDimSPCI}$ maintains valid coverage on a wide range of multivariate time series while producing smaller prediction regions than CP and non-CP baselines.

We investigate the potential of patent data for improving the antibody humanness prediction using a multi-stage, multi-loss training process. Humanness serves as a proxy for the immunogenic response to antibody therapeutics, one of the major causes of attrition in drug discovery and a challenging obstacle for their use in clinical settings. We pose the initial learning stage as a weakly-supervised contrastive-learning problem, where each antibody sequence is associated with possibly multiple identifiers of function and the objective is to learn an encoder that groups them according to their patented properties. We then freeze a part of the contrastive encoder and continue training it on the patent data using the cross-entropy loss to predict the humanness score of a given antibody sequence. We illustrate the utility of the patent data and our approach by performing inference on three different immunogenicity datasets, unseen during training. Our empirical results demonstrate that the learned model consistently outperforms the alternative baselines and establishes new state-of-the-art on five out of six inference tasks, irrespective of the used metric.

Following the introduction of Adam, several novel adaptive optimizers for deep learning have been proposed. These optimizers typically excel in some tasks but may not outperform Adam uniformly across all tasks. In this work, we introduce Meta-Adaptive Optimizers (MADA), a unified optimizer framework that can generalize several known optimizers and dynamically learn the most suitable one during training. The key idea in MADA is to parameterize the space of optimizers and dynamically search through it using hyper-gradient descent during training. We empirically compare MADA to other popular optimizers on vision and language tasks, and find that MADA consistently outperforms Adam and other popular optimizers, and is robust against sub-optimally tuned hyper-parameters. MADA achieves a greater validation performance improvement over Adam compared to other popular optimizers during GPT-2 training and fine-tuning. We also propose AVGrad, a modification of AMSGrad that replaces the maximum operator with averaging, which is more suitable for hyper-gradient optimization. Finally, we provide a convergence analysis to show that parameterized interpolations of optimizers can improve their error bounds (up to constants), hinting at an advantage for meta-optimizers.

We introduce Optimal Eye Surgeon (OES), a framework for pruning and training deep image generator networks. Typically, untrained deep convolutional networks, which include image sampling operations, serve as effective image priors. However, they tend to overfit to noise in image restoration tasks due to being overparameterized. OES addresses this by adaptively pruning networks at random initialization to a level of underparameterization. This process effectively captures low-frequency image components even without training, by just masking. When trained to fit noisy image, these pruned subnetworks, which we term Sparse-DIP, resist overfitting to noise. This benefit arises from underparameterization and the regularization effect of masking, constraining them in the manifold of image priors. We demonstrate that subnetworks pruned through OES surpass other leading pruning methods, such as the Lottery Ticket Hypothesis, which is known to be suboptimal for image recovery tasks. Our extensive experiments demonstrate the transferability of OES-masks and the characteristics of sparse-subnetworks for image generation. Code is available at https://github.com/Avra98/Optimal-Eye-Surgeon.

Pedestrian trajectory prediction aims to predict future trajectories based on observed trajectories. Current state-of-the-art methods often assume that the observed sequences of agents are complete, which is a strong assumption that overlooks inherent uncertainties. Understanding pedestrian behavior when dealing with missing values in the observed sequence is crucial for enhancing the performance of predictive models. In this work, we propose the MultiScale hypergraph for Trajectory Imputation and Prediction (MS-TIP), a novel approach that simultaneously addresses the imputation of missing observations and the prediction of future trajectories. Specifically, we leverage transformers with diagonal masked self-attention to impute incomplete observations. Further, our approach promotes complex interaction modeling through multi-scale hypergraphs, optimizing our trajectory prediction module to capture different types of interactions. With the inclusion of scenic attention, we learn contextual scene information, instead of sole reliance on coordinates. Additionally, our approach utilizes an intermediate control point and refinement module to infer future trajectories accurately. Extensive experiments validate the efficacy of MS-TIP in precisely predicting pedestrian future trajectories. Code is publicly available at https://github.com/Pranav-chib/MS-TIP.

Decision support systems for classification tasks are predominantly designed to predict the value of the ground truth labels. However, since their predictions are not perfect, these systems also need to make human experts understand when and how to use these predictions to update their own predictions. Unfortunately, this has been proven challenging. In this context, it has been recently argued that an alternative type of decision support systems may circumvent this challenge. Rather than providing a single label prediction, these systems provide a set of label prediction values constructed using a conformal predictor, namely a prediction set, and forcefully ask experts to predict a label value from the prediction set. However, the design and evaluation of these systems have so far relied on stylized expert models, questioning their promise. In this paper, we revisit the design of this type of systems from the perspective of online learning and develop a methodology that does not require, nor assumes, an expert model. Our methodology leverages the nested structure of the prediction sets provided by any conformal predictor and a natural counterfactual monotonicity assumption to achieve an exponential improvement in regret in comparison to vanilla bandit algorithms. We conduct a large-scale human subject study ($n = 2{,}751$) to compare our methodology to several competitive baselines. The results show that, for decision support systems based on prediction sets, limiting experts’ level of agency leads to greater performance than allowing experts to always exercise their own agency.

In recent years, there has been a surge in the development of 3D structure-based pre-trained protein models, representing a significant advancement over pre-trained protein language models in various downstream tasks. However, most existing structure-based pre-trained models primarily focus on the residue level, i.e., alpha carbon atoms, while ignoring other atoms like side chain atoms. We argue that modeling proteins at both residue and atom levels is important since the side chain atoms can also be crucial for numerous downstream tasks, for example, molecular docking. Nevertheless, we find that naively combining residue and atom information during pre-training typically fails. We identify a key reason is the information leakage caused by the inclusion of atom structure in the input, which renders residue-level pre-training tasks trivial and results in insufficiently expressive residue representations. To address this issue, we introduce a span mask pre-training strategy on 3D protein chains to learn meaningful representations of both residues and atoms. This leads to a simple yet effective approach to learning protein representation suitable for diverse downstream tasks. Extensive experimental results on binding site prediction and function prediction tasks demonstrate our proposed pre-training approach significantly outperforms other methods. Our code will be made public.

Current theoretical and empirical research in neural networks suggests that complex datasets require large network architectures for thorough classification, yet the precise nature of this relationship remains unclear. This paper tackles this issue by defining upper and lower bounds for neural network widths, which are informed by the polytope structure of the dataset in question. We also delve into the application of these principles to simplicial complexes and specific manifold shapes, explaining how the requirement for network width varies in accordance with the geometric complexity of the dataset. Moreover, we develop an algorithm to investigate a converse situation where the polytope structure of a dataset can be inferred from its corresponding trained neural networks. Through our algorithm, it is established that popular datasets such as MNIST, Fashion-MNIST, and CIFAR10 can be efficiently encapsulated using no more than two polytopes with a small number of faces.

The effectiveness of neural processes (NPs) in modelling posterior prediction maps---the mapping from data to posterior predictive distributions---has significantly improved since their inception. This improvement can be attributed to two principal factors: (1) advancements in the architecture of permutation invariant set functions, which are intrinsic to all NPs; and (2) leveraging symmetries present in the true posterior predictive map, which are problem dependent. Transformers are a notable development in permutation invariant set functions, and their utility within NPs has been demonstrated through the family of models we refer to as TNPs. Despite significant interest in TNPs, little attention has been given to incorporating symmetries. Notably, the posterior prediction maps for data that are stationary---a common assumption in spatio-temporal modelling---exhibit translation equivariance. In this paper, we introduce of a new family of translation equivariant TNPs that incorporate *translation equivariance*. Through an extensive range of experiments on synthetic and real-world spatio-temporal data, we demonstrate the effectiveness of TE-TNPs relative to their non-translation-equivariant counterparts and other NP baselines.

Use of machine learning to perform database operations, such as indexing, cardinality estimation, and sorting, is shown to provide substantial performance benefits. However, when datasets change and data distribution shifts, empirical results also show performance degradation for learned models, possibly to worse than non-learned alternatives. This, together with a lack of theoretical understanding of learned methods undermines their practical applicability, since there are no guarantees on how well the models will perform after deployment. In this paper, we present the first known theoretical characterization of the performance of learned models in dynamic datasets, for the aforementioned operations. Our results show novel theoretical characteristics achievable by learned models and provide bounds on the performance of the models that characterize their advantages over non-learned methods, showing why and when learned models can outperform the alternatives. Our analysis develops the *distribution learnability* framework and novel theoretical tools which build the foundation for the analysis of learned database operations in the future.

Algorithmic predictions are increasingly used to inform the allocations of goods and interventions in the public sphere. In these domains, predictions serve as a means to an end. They provide stakeholders with insights into likelihood of future events as a means to improve decision making quality, and enhance social welfare. However, if maximizing welfare is the ultimate goal, prediction is only a small piece of the puzzle. There are various other policy levers a social planner might pursue in order to improve bottom-line outcomes, such as expanding access to available goods, or increasing the effect sizes of interventions. Given this broad range of design decisions, a basic question to ask is: What is the relative value of prediction in algorithmic decision making? How do the improvements in welfare arising from better predictions compare to those of other policy levers? The goal of our work is to initiate the formal study of these questions. Our main results are theoretical in nature. We identify simple, sharp conditions determining the relative value of prediction vis-à-vis expanding access, within several statistical models that are popular amongst quantitative social scientists. Furthermore, we illustrate how these theoretical insights can guide the design of algorithmic decision making systems in practice.

Diffusion models exhibit powerful generative capabilities enabling noise mapping to data via reverse stochastic differential equations. However, in image restoration, the focus is on the mapping relationship from low-quality to high-quality images. Regarding this issue, we introduce the Generalized Ornstein-Uhlenbeck Bridge (GOUB) model. By leveraging the natural mean-reverting property of the generalized OU process and further eliminating the variance of its steady-state distribution through the Doob's *h*–transform, we achieve diffusion mappings from point to point enabling the recovery of high-quality images from low-quality ones. Moreover, we unravel the fundamental mathematical essence shared by various bridge models, all of which are special instances of GOUB and empirically demonstrate the optimality of our proposed models. Additionally, we present the corresponding Mean-ODE model adept at capturing both pixel-level details and structural perceptions. Experimental outcomes showcase the state-of-the-art performance achieved by both models across diverse tasks, including inpainting, deraining, and super-resolution. Code is available at https://github.com/Hammour-steak/GOUB.

Despite their dominance in modern DL and, especially, NLP domains, transformer architectures exhibit sub-optimal performance on long-range tasks compared to recent layers that are specifically designed for this purpose. In this work, drawing inspiration from key attributes of longrange layers, such as state-space layers, linear RNN layers, and global convolution layers, we demonstrate that minimal modifications to the transformer architecture can significantly enhance performance on the Long Range Arena (LRA) benchmark, thus narrowing the gap with these specialized layers. We identify that two key principles for long-range tasks are (i) incorporating an inductive bias towards smoothness, and (ii) locality. As we show, integrating these ideas into the attention mechanism improves results with a negligible amount of additional computation and without any additional trainable parameters. Our theory and experiments also shed light on the reasons for the inferior performance of transformers on long-range tasks and identify critical properties that are essential for successfully capturing long-range dependencies.

In this paper, we investigate the long-term memory learning capabilities of state-space models (SSMs) from the perspective of parameterization. We prove that state-space models without any reparameterization exhibit a memory limitation similar to that of traditional RNNs: the target relationships that can be stably approximated by state-space models must have an exponential decaying memory. Our analysis identifies this ``curse of memory'' as a result of the recurrent weights converging to a stability boundary, suggesting that a reparameterization technique can be effective. To this end, we introduce a class of reparameterization techniques for SSMs that effectively lift its memory limitations. Besides improving approximation capabilities, we further illustrate that a principled choice of reparameterization scheme can also enhance optimization stability. We validate our findings using synthetic datasets, language models and image classifications.

Research into optimisation for deep learning is characterised by a tension between the computational efficiency of first-order, gradient-based methods (such as SGD and Adam) and the theoretical efficiency of second-order, curvature-based methods (such as quasi-Newton methods and K-FAC). Noting that second-order methods often only function effectively with the addition of stabilising heuristics (such as Levenberg-Marquardt damping), we ask how much these (as opposed to the second-order curvature model) contribute to second-order algorithms' performance. We thus study _AdamQLR_: an optimiser combining damping and learning rate selection techniques from K-FAC (Martens & Grosse, 2015) with the update directions proposed by Adam, inspired by considering Adam through a second-order lens. We evaluate AdamQLR on a range of regression and classification tasks at various scales and hyperparameter tuning methodologies, concluding K-FAC's adaptive heuristics are of variable standalone general effectiveness, and finding an _untuned_ AdamQLR setting can achieve comparable performance vs runtime to _tuned_ benchmarks.

Approaches to predicting neuronal spike responses commonly use a Poisson learning objective. This objective quantizes responses into spike counts within a fixed summation interval, typically on the order of 10 to 100 milliseconds in duration; however, neuronal responses are often time accurate down to a few milliseconds, and Poisson models struggle to precisely model them at these timescales. We propose the concept of a spike distance function that maps points in time to the temporal distance to the nearest spike. We show that neural networks can be trained to approximate spike distance functions, and we present an efficient algorithm for inferring spike trains from the outputs of these models. Using recordings of chicken and frog retinal ganglion cells responding to visual stimuli, we compare the performance of our approach to that of Poisson models trained with various summation intervals. We show that our approach outperforms the use of Poisson models at spike train inference.

Fluid prediction is a long-standing challenge due to the intrinsic high-dimensional non-linear dynamics. Previous methods usually utilize the non-linear modeling capability of deep models to directly estimate velocity fields for future prediction. However, skipping over inherent physical properties but directly learning superficial velocity fields will overwhelm the model from generating precise or physics-reliable results. In this paper, we propose the HelmFluid toward an accurate and interpretable predictor for fluid. Inspired by the Helmholtz theorem, we design a HelmDynamics block to learn Helmholtz dynamics, which decomposes fluid dynamics into more solvable curl-free and divergence-free parts, physically corresponding to potential and stream functions of fluid. By embedding the HelmDynamics block into a Multiscale Multihead Integral Architecture, HelmFluid can integrate learned Helmholtz dynamics along temporal dimension in multiple spatial scales to yield future fluid. Compared with previous velocity estimating methods, HelmFluid is faithfully derived from Helmholtz theorem and ravels out complex fluid dynamics with physically interpretable evidence. Experimentally, HelmFluid achieves consistent state-of-the-art in both numerical simulated and real-world observed benchmarks, even for scenarios with complex boundaries.