NeurIPS 2022 Papers

AI-empowered drug recommendation has become an important task in healthcare research areas, which offers an additional perspective to assist human doctors with more accurate and more efficient drug prescriptions. Generally, drug recommendation is based on patients' diagnosis results in the electronic health records. We assume that there are three key factors to be addressed in drug recommendation: 1) elimination of recommendation bias due to limitations of observable information, 2) better utilization of historical health condition and 3) coordination of multiple drugs to control safety. To this end, we propose DrugRec, a causal inference based drug recommendation model. The causal graphical model can identify and deconfound the recommendation bias with front-door adjustment. Meanwhile, we model the multi-visit in the causal graph to characterize a patient's historical health conditions. Finally, we model the drug-drug interactions (DDIs) as the propositional satisfiability (SAT) problem, and solving the SAT problem can help better coordinate the recommendation. Comprehensive experiment results show that our proposed model achieves state-of-the-art performance on the widely used datasets MIMIC-III and MIMIC-IV, demonstrating the effectiveness and safety of our method.

Generative models for graph data are an important research topic in machine learning. Graph data comprise two levels that are typically analyzed separately: node-level properties such as the existence of a link between a pair of nodes, and global aggregate graph-level statistics, such as motif counts. This paper proposes a new multi-level framework that jointly models node-level properties and graph-level statistics, as mutually reinforcing sources of information. We introduce a new micro-macro training objective for graph generation that combines node-level and graph-level losses. We utilize the micro-macro objective to improve graph generation with a GraphVAE, a well-established model based on graph-level latent variables, that provides fast training and generation time for medium-sized graphs. Our experiments show that adding micro-macro modeling to the GraphVAE model improves graph quality scores up to 2 orders of magnitude on five benchmark datasets, while maintaining the GraphVAE generation speed advantage.

Recent studies have shown that deep learning models such as RNNs and Transformers have brought significant performance gains for long-term forecasting of time series because they effectively utilize historical information. We found, however, that there is still great room for improvement in how to preserve historical information in neural networks while avoiding overfitting to noise present in the history. Addressing this allows better utilization of the capabilities of deep learning models. To this end, we design a Frequency improved Legendre Memory model, or FiLM: it applies Legendre polynomial projections to approximate historical information, uses Fourier projection to remove noise, and adds a low-rank approximation to speed up computation. Our empirical studies show that the proposed FiLM significantly improves the accuracy of state-of-the-art models in multivariate and univariate long-term forecasting by (19.2%, 22.6%), respectively. We also demonstrate that the representation module developed in this work can be used as a general plugin to improve the long-term prediction performance of other deep learning modules. Code is available at https://github.com/tianzhou2011/FiLM/.

The ability to extrapolate from short problem instances to longer ones is an important form of out-of-distribution generalization in reasoning tasks, and is crucial when learning from datasets where longer problem instances are rare. These include theorem proving, solving quantitative mathematics problems, and reading/summarizing novels. In this paper, we run careful empirical studies exploring the length generalization capabilities of transformer-based language models. We first establish that naively finetuning transformers on length generalization tasks shows significant generalization deficiencies independent of model scale. We then show that combining pretrained large language models' in-context learning abilities with scratchpad prompting (asking the model to output solution steps before producing an answer) results in a dramatic improvement in length generalization. We run careful failure analyses on each of the learning modalities and identify common sources of mistakes that highlight opportunities in equipping language models with the ability to generalize to longer problems.

Planning enables autonomous agents to solve complex decision-making problems by evaluating predictions of the future. However, classical planning algorithms often become infeasible in real-world settings where state spaces are high-dimensional and transition dynamics unknown. The idea behind latent planning is to simplify the decision-making task by mapping it to a lower-dimensional embedding space. Common latent planning strategies are based on trajectory optimization techniques such as shooting or collocation, which are prone to failure in long-horizon and highly non-convex settings. In this work, we study long-horizon goal-reaching scenarios from visual inputs and formulate latent planning as an explorative tree search. Inspired by classical sampling-based motion planning algorithms, we design a method which iteratively grows and optimizes a tree representation of visited areas of the latent space. To encourage fast exploration, the sampling of new states is biased towards sparsely represented regions within the estimated data support. Our method, called Expansive Latent Space Trees (ELAST), relies on self-supervised training via contrastive learning to obtain (a) a latent state representation and (b) a latent transition density model. We embed ELAST into a model-predictive control scheme and demonstrate significant performance improvements compared to existing baselines given challenging visual control tasks in simulation, including the navigation for a deformable object.

Contrastive learning, which learns to contrast positive with negative pairs of samples, has been popular for self-supervised visual representation learning. Although great effort has been made to design proper positive pairs through data augmentation, few works attempt to generate optimal positives for each instance. Inspired by semantic consistency and computational advantage in latent space of pretrained generative models, this paper proposes to learn instance-specific latent transformations to generate Contrastive Optimal Positives (COP-Gen) for self-supervised contrastive learning. Specifically, we formulate COP-Gen as an instance-specific latent space navigator which minimizes the mutual information between the generated positive pair subject to the semantic consistency constraint. Theoretically, the learned latent transformation creates optimal positives for contrastive learning, which removes as much nuisance information as possible while preserving the semantics. Empirically, using generated positives by COP-Gen consistently outperforms other latent transformation methods and even real-image-based methods in self-supervised contrastive learning.

We study the problem of training deep networks while quantizing parameters and activations into low-precision numeric representations, a setting central to reducing energy consumption and inference time of deployed models. We propose a method that learns different precisions, as measured by bits in numeric representations, for different weights in a neural network, yielding a heterogeneous allocation of bits across parameters. Learning precisions occurs alongside learning weight values, using a strategy derived from a novel framework wherein the intractability of optimizing discrete precisions is approximated by training per-parameter noise magnitudes. We broaden this framework to also encompass learning precisions for hidden state activations, simultaneously with weight precisions and values. Our approach exposes the objective of constructing a low-precision inference-efficient model to the entirety of the training process. Experiments show that it finds highly heterogeneous precision assignments for CNNs trained on CIFAR and ImageNet, improving upon previous state-of-the-art quantization methods. Our improvements extend to the challenging scenario of learning reduced-precision GANs.

Understanding genetic variation, e.g., through mutations, in organisms is crucial to unravel their effects on the environment and human health. A fundamental characterization can be obtained by solving the haplotype assembly problem, which yields the variation across multiple copies of chromosomes. Variations among fast evolving viruses that lead to different strains (called quasispecies) are also deciphered with similar approaches. In both these cases, high-throughput sequencing technologies that provide oversampled mixtures of large noisy fragments (reads) of genomes, are used to infer constituent components (haplotypes or quasispecies). The problem is harder for polyploid species where there are more than two copies of chromosomes. State-of-the-art neural approaches to solve this NP-hard problem do not adequately model relations among the reads that are important for deconvolving the input signal. We address this problem by developing a new method, called NeurHap, that combines graph representation learning with combinatorial optimization. Our experiments demonstrate the substantially better performance of NeurHap in real and synthetic datasets compared to competing approaches.

From optimal transport to robust dimensionality reduction, many machine learning applications can be cast into the min-max optimization problems over Riemannian manifolds. Though many min-max algorithms have been analyzed in the Euclidean setting, it has been elusive how these results translate to the Riemannian case. Zhang et al. (2022) have recently identified that geodesic convex concave Riemannian problems admit always Sion’s saddle point solutions. Immediately, an important question that arises is if a performance gap between the Riemannian and the optimal Euclidean space convex concave algorithms is necessary. Our work is the first to answer the question in the negative: We prove that the Riemannian corrected extragradient (RCEG) method achieves last-iterate at a linear convergence rate at the geodesically strongly convex concave case, matching the euclidean one. Our results also extend to the stochastic or non-smooth case where RCEG & Riemanian gradient ascent descent (RGDA) achieve respectively near-optimal convergence rates up to factors depending on curvature of the manifold. Finally, we empirically demonstrate the effectiveness of RCEG in solving robust PCA.

Recent breakthroughs in the development of agents to solve challenging sequential decision making problems such as Go, StarCraft, or DOTA, have relied on both simulated environments and large-scale datasets. However, progress on this research has been hindered by the scarcity of open-sourced datasets and the prohibitive computational cost to work with them. Here we present the NetHack Learning Dataset (NLD), a large and highly-scalable dataset of trajectories from the popular game of NetHack, which is both extremely challenging for current methods and very fast to run. NLD consists of three parts: 10 billion state transitions from 1.5 million human trajectories collected on the NAO public NetHack server from 2009 to 2020; 3 billion state-action-score transitions from 100,000 trajectories collected from the symbolic bot winner of the NetHack Challenge 2021; and, accompanying code for users to record, load and stream any collection of such trajectories in a highly compressed form. We evaluate a wide range of existing algorithms for learning from demonstrations, showing that significant research advances are needed to fully leverage large-scale datasets for challenging sequential decision making tasks.

On-device training enables the model to adapt to new data collected from the sensors by fine-tuning a pre-trained model. Users can benefit from customized AI models without having to transfer the data to the cloud, protecting the privacy. However, the training memory consumption is prohibitive for IoT devices that have tiny memory resources. We propose an algorithm-system co-design framework to make on-device training possible with only 256KB of memory. On-device training faces two unique challenges: (1) the quantized graphs of neural networks are hard to optimize due to low bit-precision and the lack of normalization; (2) the limited hardware resource (memory and computation) does not allow full backpropagation. To cope with the optimization difficulty, we propose Quantization- Aware Scaling to calibrate the gradient scales and stabilize 8-bit quantized training. To reduce the memory footprint, we propose Sparse Update to skip the gradient computation of less important layers and sub-tensors. The algorithm innovation is implemented by a lightweight training system, Tiny Training Engine, which prunes the backward computation graph to support sparse updates and offload the runtime auto-differentiation to compile time. Our framework is the first practical solution for on-device transfer learning of visual recognition on tiny IoT devices (e.g., a microcontroller with only 256KB SRAM), using less than 1/1000 of the memory of PyTorch and TensorFlow while matching the accuracy. Our study enables IoT devices not only to perform inference but also to continuously adapt to new data for on-device lifelong learning. A video demo can be found here: https://youtu.be/XaDCO8YtmBw.

Diffusion-based generative models are extremely effective in generating high-quality images, with generated samples often surpassing the quality of those produced by other models under several metrics. One distinguishing feature of these models, however, is that they typically require long sampling chains in order to produce high-fidelity images. This presents a challenge not only from the lenses of sampling time, but also from the inherent difficulty in backpropagating through these chains in order to accomplish tasks such as model inversion, i.e., approximately finding latent states that generate known images. In this paper, we look at diffusion models through a different perspective, that of a (deep) equilibrium (DEQ) fixed point model. Specifically, we extend the recent denoising diffusion implicit model (DDIM), and model the entire sampling chain as a joint, multi-variate fixed point system. This setup provides an elegant unification of diffusion and equilibrium models, and shows benefits in 1) single-shot image sampling, as it replaces the fully-serial typical sampling process with a parallel one; and 2) model inversion, where we can leverage fast gradients in the DEQ setting to much more quickly find the noise that generates a given image. The approach is also orthogonal and thus complementary to other methods used to reduce the sampling time, or improve model inversion. We demonstrate our method's strong performance across several datasets, including CIFAR10, CelebA, and LSUN Bedroom and Churches.

In many real-world cooperative multiagent reinforcement learning (MARL) tasks, teams of agents can rehearse together before deployment, but then communication constraints may force individual agents to execute independently when deployed. Centralized training and decentralized execution (CTDE) is increasingly popular in recent years, focusing mainly on this setting. In the value-based MARL branch, credit assignment mechanism is typically used to factorize the team reward into each individual’s reward — individual-global-max (IGM) is a condition on the factorization ensuring that agents’ action choices coincide with team’s optimal joint action. However, current architectures fail to consider local coordination within sub-teams that should be exploited for more effective factorization, leading to faster learning. We propose a novel value factorization framework, called multiagent Q-learning with sub-team coordination (QSCAN), to flexibly represent sub-team coordination while honoring the IGM condition. QSCAN encompasses the full spectrum of sub-team coordination according to sub-team size, ranging from the monotonic value function class to the entire IGM function class, with familiar methods such as QMIX and QPLEX located at the respective extremes of the spectrum. Experimental results show that QSCAN’s performance dominates state-of-the-art methods in matrix games, predator-prey tasks, the Switch challenge in MA-Gym. Additionally, QSCAN achieves comparable performances to those methods in a selection of StarCraft II micro-management tasks.

Formal XAI (explainable AI) is a growing area that focuses on computing explanations with mathematical guarantees for the decisions made by ML models. Inside formal XAI, one of the most studied cases is that of explaining the choices taken by decision trees, as they are traditionally deemed as one of the most interpretable classes of models. Recent work has focused on studying the computation of sufficient reasons, a kind of explanation in which given a decision tree $T$ and an instance $x$, one explains the decision $T(x)$ by providing a subset $y$ of the features of $x$ such that for any other instance $z$ compatible with $y$, it holds that $T(z) = T(x)$, intuitively meaning that the features in $y$ are already enough to fully justify the classification of $x$ by $T$. It has been argued, however, that sufficient reasons constitute a restrictive notion of explanation. For such a reason, the community has started to study their probabilistic counterpart, in which one requires that the probability of $T(z) = T(x)$ must be at least some value $\delta \in (0, 1]$, where $z$ is a random instance that is compatible with $y$. Our paper settles the computational complexity of $\delta$-sufficient-reasons over decision trees, showing that both (1) finding $\delta$-sufficient-reasons that are minimal in size, and (2) finding $\delta$-sufficient-reasons that are minimal inclusion-wise, do not admit polynomial-time algorithms (unless P = NP). This is in stark contrast with the deterministic case ($\delta = 1$) where inclusion-wise minimal sufficient-reasons are easy to compute. By doing this, we answer two open problems originally raised by Izza et al., and extend the hardness of explanations for Boolean circuits presented by W{\"a}ldchen et al. to the more restricted case of decision trees. On the positive side, we identify structural restrictions of decision trees that make the problem tractable, and show how SAT solvers might be able to tackle these problems in practical settings.

Column Generation (CG) is an iterative algorithm for solving linear programs (LPs) with an extremely large number of variables (columns). CG is the workhorse for tackling large-scale integer linear programs, which rely on CG to solve LP relaxations within a branch and bound algorithm. Two canonical applications are the Cutting Stock Problem (CSP) and Vehicle Routing Problem with Time Windows (VRPTW). In VRPTW, for example, each binary variable represents the decision to include or exclude a route, of which there are exponentially many; CG incrementally grows the subset of columns being used, ultimately converging to an optimal solution. We propose RLCG, the first Reinforcement Learning (RL) approach for CG. Unlike typical column selection rules which myopically select a column based on local information at each iteration, we treat CG as a sequential decision-making problem, as the column selected in an iteration affects subsequent iterations of the algorithm. This perspective lends itself to a Deep Reinforcement Learning approach that uses Graph Neural Networks (GNNs) to represent the variable-constraint structure in the LP of interest. We perform an extensive set of experiments using the publicly available BPPLIB benchmark for CSP and Solomon benchmark for VRPTW. RLCG converges faster and reduces the number of CG iterations by 22.4% for CSP and 40.9% for VRPTW on average compared to a commonly used greedy policy.

There are over 300 sign languages in the world, many of which have very limited or no labelled sign-to-text datasets. To address low-resource data scenarios, self-supervised pretraining and multilingual finetuning have been shown to be effective in natural language and speech processing. In this work, we apply these ideas to sign language recognition. We make three contributions. - First, we release SignCorpus, a large pretraining dataset on sign languages comprising about 4.6K hours of signing data across 10 sign languages. SignCorpus is curated from sign language videos on the internet, filtered for data quality, and converted into sequences of pose keypoints thereby removing all personal identifiable information (PII). - Second, we release Sign2Vec, a graph-based model with 5.2M parameters that is pretrained on SignCorpus. We envisage Sign2Vec as a multilingual large-scale pretrained model which can be fine-tuned for various sign recognition tasks across languages. - Third, we create MultiSign-ISLR -- a multilingual and label-aligned dataset of sequences of pose keypoints from 11 labelled datasets across 7 sign languages, and MultiSign-FS -- a new finger-spelling training and test set across 7 languages. On these datasets, we fine-tune Sign2Vec to create multilingual isolated sign recognition models. With experiments on multiple benchmarks, we show that pretraining and multilingual transfer are effective giving significant gains over state-of-the-art results. All datasets, models, and code has been made open-source via the OpenHands toolkit.

Partially monotone regression is a regression analysis in which the target values are monotonically increasing with respect to a subset of input features. The TensorFlow Lattice library is one of the standard machine learning libraries for partially monotone regression. It consists of several neural network layers, and its core component is the lattice layer. One of the problems of the lattice layer is that it requires the projected gradient descent algorithm with many constraints to train it. Another problem is that it cannot receive a high-dimensional input vector due to the memory consumption. We propose a novel neural network layer, the hierarchical lattice layer (HLL), as an extension of the lattice layer so that we can use a standard stochastic gradient descent algorithm to train HLL while satisfying monotonicity constraints and so that it can receive a high-dimensional input vector. Our experiments demonstrate that HLL did not sacrifice its prediction performance on real datasets compared with the lattice layer.

This work identifies the existence and cause of a type of posterior collapse that frequently occurs in the Bayesian deep learning practice. For a general linear latent variable model that includes linear variational autoencoders as a special case, we precisely identify the nature of posterior collapse to be the competition between the likelihood and the regularization of the mean due to the prior. Our result also suggests that posterior collapse may be a general problem of learning for deeper architectures and deepens our understanding of Bayesian deep learning.

We analyze the dynamics of large batch stochastic gradient descent with momentum (SGD+M) on the least squares problem when both the number of samples and dimensions are large. In this setting, we show that the dynamics of SGD+M converge to a deterministic discrete Volterra equation as dimension increases, which we analyze. We identify a stability measurement, the implicit conditioning ratio (ICR), which regulates the ability of SGD+M to accelerate the algorithm. When the batch size exceeds this ICR, SGD+M converges linearly at a rate of $\mathcal{O}(1/\sqrt{\kappa})$, matching optimal full-batch momentum (in particular performing as well as a full-batch but with a fraction of the size). For batch sizes smaller than the ICR, in contrast, SGD+M has rates that scale like a multiple of the single batch SGD rate. We give explicit choices for the learning rate and momentum parameter in terms of the Hessian spectra that achieve this performance.