We present a noisy channel generative model of two sequences, for example text and speech, which enables uncovering the association between the two modalities when limited paired data is available. To address the intractability of the exact model under a realistic data setup, we propose a variational inference approximation. To train this variational model with categorical data, we propose a KL encoder loss approach which has connections to the wake-sleep algorithm. Identifying the joint or conditional distributions by only observing unpaired samples from the marginals is only possible under certain conditions in the data distribution and we discuss under what type of conditional independence assumptions that might be achieved, which guides the architecture designs. Experimental results show that even tiny amount of paired data (5 minutes) is sufficient to learn to relate the two modalities (graphemes and phonemes here) when a massive amount of unpaired data is available, paving the path to adopting this principled approach for all seq2seq models in low data resource regimes.
相關內容
Stochastic human motion prediction (HMP) has generally been tackled with generative adversarial networks and variational autoencoders. Most prior works aim at predicting highly diverse movements in terms of the skeleton joints' dispersion. This has led to methods predicting fast and motion-divergent movements, which are often unrealistic and incoherent with past motion. Such methods also neglect contexts that need to anticipate diverse low-range behaviors, or actions, with subtle joint displacements. To address these issues, we present BeLFusion, a model that, for the first time, leverages latent diffusion models in HMP to sample from a latent space where behavior is disentangled from pose and motion. As a result, diversity is encouraged from a behavioral perspective. Thanks to our behavior coupler's ability to transfer sampled behavior to ongoing motion, BeLFusion's predictions display a variety of behaviors that are significantly more realistic than the state of the art. To support it, we introduce two metrics, the Area of the Cumulative Motion Distribution, and the Average Pairwise Distance Error, which are correlated to our definition of realism according to a qualitative study with 126 participants. Finally, we prove BeLFusion's generalization power in a new cross-dataset scenario for stochastic HMP.
Inverse Models for Estimating the Initial Condition of Spatio-Temporal Advection-Diffusion Processes
Inverse problems involve making inference about unknown parameters of a physical process using observational data. This paper investigates an important class of inverse problems -- the estimation of the initial condition of a spatio-temporal advection-diffusion process using spatially sparse data streams. Three spatial sampling schemes are considered, including irregular, non-uniform and shifted uniform sampling. The irregular sampling scheme is the general scenario, while computationally efficient solutions are available in the spectral domain for non-uniform and shifted uniform sampling. For each sampling scheme, the inverse problem is formulated as a regularized convex optimization problem that minimizes the distance between forward model outputs and observations. The optimization problem is solved by the Alternating Direction Method of Multipliers algorithm, which also handles the situation when a linear inequality constraint (e.g., non-negativity) is imposed on the model output. Numerical examples are presented, code is made available on GitHub, and discussions are provided to generate some useful insights of the proposed inverse modeling approaches.
Estimation of heterogeneous causal effects - i.e., how effects of policies and treatments vary across subjects - is a fundamental task in causal inference, playing a crucial role in optimal treatment allocation, generalizability, subgroup effects, and more. Many flexible methods for estimating conditional average treatment effects (CATEs) have been proposed in recent years, but questions surrounding optimality have remained largely unanswered. In particular, a minimax theory of optimality has yet to be developed, with the minimax rate of convergence and construction of rate-optimal estimators remaining open problems. In this paper we derive the minimax rate for CATE estimation, in a nonparametric model where distributional components are Holder-smooth, and present a new local polynomial estimator, giving high-level conditions under which it is minimax optimal. More specifically, our minimax lower bound is derived via a localized version of the method of fuzzy hypotheses, combining lower bound constructions for nonparametric regression and functional estimation. Our proposed estimator can be viewed as a local polynomial R-Learner, based on a localized modification of higher-order influence function methods; it is shown to be minimax optimal under a condition on how accurately the covariate distribution is estimated. The minimax rate we find exhibits several interesting features, including a non-standard elbow phenomenon and an unusual interpolation between nonparametric regression and functional estimation rates. The latter quantifies how the CATE, as an estimand, can be viewed as a regression/functional hybrid. We conclude with some discussion of a few remaining open problems.
Pre-trained models have been used in many fields in recent years, ranging from natural language understanding to computer vision and natural language generation. However, the performance of these natural language generation models is overly dependent on the scale of the model and the size of the dataset. While the larger language model is excellent in some respects, it cannot learn up-to-date knowledge and is relatively difficult to relearn. In this paper, a new adversarial process learning method called Auto-Learning. This can improve the performance of any natural language generation model without the help of additional datasets. Auto-Learning includes two models: $G$ is a text generation model and $D$ can test whether the data generated by G is legitimate. Firstly, the fine-tuned $D$ model is used as the brain's knowledge base before the process. Then the text generated by the $G$ model is used as the input of $D$ to determine whether the text is legitimate or not. Finally, $G$ is fine-tuned according to the output of $D$. This adversarial process is like a self-escalation of the brain through some a priori knowledge. When this adversarial system wants to learn something new, simply fine-tune the $D$ model. Our approach applies to Autoregressive Language Modeling for all Transformer classes. The results are good in existing experimental tasks, including more grammatical text generation and better performance on some text comprehension tasks.
A confidence sequence (CS) is a sequence of confidence intervals that is valid at arbitrary data-dependent stopping times. These are useful in applications like A/B testing, multi-armed bandits, off-policy evaluation, election auditing, etc. We present three approaches to constructing a confidence sequence for the population mean, under the minimal assumption that only an upper bound $\sigma^2$ on the variance is known. While previous works rely on light-tail assumptions like boundedness or subGaussianity (under which all moments of a distribution exist), the confidence sequences in our work are able to handle data from a wide range of heavy-tailed distributions. The best among our three methods -- the Catoni-style confidence sequence -- performs remarkably well in practice, essentially matching the state-of-the-art methods for $\sigma^2$-subGaussian data, and provably attains the $\sqrt{\log \log t/t}$ lower bound due to the law of the iterated logarithm. Our findings have important implications for sequential experimentation with unbounded observations, since the $\sigma^2$-bounded-variance assumption is more realistic and easier to verify than $\sigma^2$-subGaussianity (which implies the former). We also extend our methods to data with infinite variance, but having $p$-th central moment ($1<p<2$).
Research on debiased recommendation has shown promising results. However, some issues still need to be handled for its application in industrial recommendation. For example, most of the existing methods require some specific data, architectures and training methods. In this paper, we first argue through an online study that arbitrarily removing all the biases in industrial recommendation may not consistently yield a desired performance improvement. For the situation that a randomized dataset is not available, we propose a novel self-sampling training and evaluation (SSTE) framework to achieve the accuracy-bias tradeoff in recommendation, i.e., eliminate the harmful biases and preserve the beneficial ones. Specifically, SSTE uses a self-sampling module to generate some subsets with different degrees of bias from the original training and validation data. A self-training module infers the beneficial biases and learns better tradeoff based on these subsets, and a self-evaluation module aims to use these subsets to construct more plausible references to reflect the optimized model. Finally, we conduct extensive offline experiments on two datasets to verify the effectiveness of our SSTE. Moreover, we deploy our SSTE in homepage recommendation of a famous financial management product called Tencent Licaitong, and find very promising results in an online A/B test.
In Bayesian analysis, the selection of a prior distribution is typically done by considering each parameter in the model. While this can be convenient, in many scenarios it may be desirable to place a prior on a summary measure of the model instead. In this work, we propose a prior on the model fit, as measured by a Bayesian coefficient of determination (R2), which then induces a prior on the individual parameters. We achieve this by placing a beta prior on R2 and then deriving the induced prior on the global variance parameter for generalized linear mixed models. We derive closed-form expressions in many scenarios and present several approximation strategies when an analytic form is not possible and/or to allow for easier computation. In these situations, we suggest approximating the prior by using a generalized beta prime distribution and provide a simple default prior construction scheme. This approach is quite flexible and can be easily implemented in standard Bayesian software. Lastly, we demonstrate the performance of the method on simulated data, where it particularly shines in high-dimensional examples, as well as real-world data, which shows its ability to model spatial correlation in the random effects.
We consider a network consisting of $n$ nodes that aim to track a continually updating process or event. To disseminate updates about the event to the network, two sources are available, such that information obtained from one source is considered more reliable than the other source. The nodes wish to have access to information about the event that is not only latest but also more reliable, and prefer a reliable packet over an unreliable packet even when the former is a bit outdated with respect to the latter. We study how such preference affects the fraction of users with reliable information in the network and their version age of information. We derive the analytical equations to characterize the two quantities, long-term expected fraction of nodes with reliable packets and their long-term expected version age using stochastic hybrid systems (SHS) modelling and study their properties. We also compare these results with the case where nodes give more preference to freshness of information than its reliability. Finally we show simulation results to verify the theoretical results and shed further light on behavior of above quantities with respect to dependent variables.
The integration of discrete algorithmic components in deep learning architectures has numerous applications. Recently, Implicit Maximum Likelihood Estimation (IMLE, Niepert, Minervini, and Franceschi 2021), a class of gradient estimators for discrete exponential family distributions, was proposed by combining implicit differentiation through perturbation with the path-wise gradient estimator. However, due to the finite difference approximation of the gradients, it is especially sensitive to the choice of the finite difference step size, which needs to be specified by the user. In this work, we present Adaptive IMLE (AIMLE), the first adaptive gradient estimator for complex discrete distributions: it adaptively identifies the target distribution for IMLE by trading off the density of gradient information with the degree of bias in the gradient estimates. We empirically evaluate our estimator on synthetic examples, as well as on Learning to Explain, Discrete Variational Auto-Encoders, and Neural Relational Inference tasks. In our experiments, we show that our adaptive gradient estimator can produce faithful estimates while requiring orders of magnitude fewer samples than other gradient estimators.
We present prompt distribution learning for effectively adapting a pre-trained vision-language model to address downstream recognition tasks. Our method not only learns low-bias prompts from a few samples but also captures the distribution of diverse prompts to handle the varying visual representations. In this way, we provide high-quality task-related content for facilitating recognition. This prompt distribution learning is realized by an efficient approach that learns the output embeddings of prompts instead of the input embeddings. Thus, we can employ a Gaussian distribution to model them effectively and derive a surrogate loss for efficient training. Extensive experiments on 12 datasets demonstrate that our method consistently and significantly outperforms existing methods. For example, with 1 sample per category, it relatively improves the average result by 9.1% compared to human-crafted prompts.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191