We combine the capacity of sparsely gated Mixture-of-Experts (MoE) with the speed and stability of linear, mixing transformations to design the Sparse Mixer encoder model. Sparse Mixer slightly outperforms (<1%) BERT on GLUE and SuperGLUE, but more importantly trains 65% faster and runs inference 61% faster. We also present a faster variant, prosaically named Fast Sparse Mixer, that marginally underperforms BERT on SuperGLUE, but trains and runs nearly twice as fast. We justify the design of these two models by carefully ablating through various mixing mechanisms, MoE configurations and hyperparameters. Sparse Mixer overcomes many of the latency and stability concerns of MoE models and offers the prospect of serving sparse student models, without resorting to distilling them to dense variants.

Massive multiple input multiple output (MIMO) systems are typically designed under the assumption of linear power amplifiers (PAs). However, PAs are typically most energy-efficient when operating close to their saturation point, where they cause non-linear distortion. Moreover, when using conventional precoders, this distortion coherently combines at the user locations, limiting performance. As such, when designing an energy-efficient massive MIMO system, this distortion has to be managed. In this work, we propose the use of a neural network (NN) to learn the mapping between the channel matrix and the precoding matrix, which maximizes the sum rate in the presence of this non-linear distortion. This is done for a third-order polynomial PA model for both the single and multi-user case. By learning this mapping a significant increase in energy efficiency is achieved as compared to conventional precoders and even as compared to perfect digital pre-distortion (DPD), in the saturation regime.

The normalizing layer has become one of the basic configurations of deep learning models, but it still suffers from computational inefficiency, interpretability difficulties, and low generality. After gaining a deeper understanding of the recent normalization and normalizer-free research works from a sample's perspective, we reveal the fact that the problem lies in the sampling noise and the inappropriate prior assumption. In this paper, we propose a simple and effective alternative to normalization, which is called "NoMorelization". NoMorelization is composed of two trainable scalars and a zero-centered noise injector. Experimental results demonstrate that NoMorelization is a general component for deep learning and is suitable for different model paradigms (e.g., convolution-based and attention-based models) to tackle different tasks (e.g., discriminative and generative tasks). Compared with existing mainstream normalizers (e.g., BN, LN, and IN) and state-of-the-art normalizer-free methods, NoMorelization shows the best speed-accuracy trade-off.

Cooking is a uniquely human endeavor for transforming raw ingredients into delicious dishes. Over centuries, cultures worldwide have evolved diverse cooking practices ingrained in their culinary traditions. Recipes, thus, are cultural capsules that capture culinary knowledge in elaborate cooking protocols. While simple quantitative models have probed the patterns in recipe composition and the process of cuisine evolution, unlike other cultural quirks such as language, the principles of cooking remain hitherto unexplored. The fundamental rules that drive the act of cooking, shaping recipe composition and cuisine architecture, are unclear. Here we present a generative grammar of cooking that captures the underlying culinary logic. By studying an extensive repository of structured recipes, we identify core concepts and rules that together forge a combinatorial system for culinary synthesis. Building on the body of work done in the context of language, the demonstration of a logically consistent generative framework offers profound insights into the act of cooking. Given the central role of food in nutrition and lifestyle disorders, culinary grammar provides leverage to improve public health through dietary interventions beyond applications for creative pursuits such as novel recipe generation.

Ordinary Differential Equations (ODEs) have recently gained a lot of attention in machine learning. However, the theoretical aspects, e.g., identifiability and asymptotic properties of statistical estimation are still obscure. This paper derives a sufficient condition for the identifiability of homogeneous linear ODE systems from a sequence of equally-spaced error-free observations sampled from a single trajectory. When observations are disturbed by measurement noise, we prove that under mild conditions, the parameter estimator based on the Nonlinear Least Squares (NLS) method is consistent and asymptotic normal with $n^{-1/2}$ convergence rate. Based on the asymptotic normality property, we construct confidence sets for the unknown system parameters and propose a new method to infer the causal structure of the ODE system, i.e., inferring whether there is a causal link between system variables. Furthermore, we extend the results to degraded observations, including aggregated and time-scaled ones. To the best of our knowledge, our work is the first systematic study of the identifiability and asymptotic properties in learning linear ODE systems. We also construct simulations with various system dimensions to illustrate the established theoretical results.

Then, we implement the physics-informed neural network (PiNN) on both simulated and real-world data. The PiNN model enables robust analysis of the dynamic spread, prediction, and parameter optimization of the COVID-19 compartmental models. The models exhibit relative root mean square error (RRMSE) of < 4% for all components and provide incubation, death, and recovery rates of γ = 0.0130, λ = 0.0001, and ρ = 0.0037, respectively, for the first 310 days of the epidemic in the US with RRMSE of < 0.35% for all components. To further improve the model performance, temporally varying parameters can be included, such as vaccination, transmission, and incubation rates. Our implementation highlights PiNN as a reliable candidate approach for forecasting real-world data and can be applied to other compartmental model variants of interest.

Leveraging task-aware annotated data as supervised signals to assist with self-supervised learning on large-scale unlabeled data has become a new trend in pre-training language models. Existing studies show that multi-task learning with large-scale supervised tasks suffers from negative effects across tasks. To tackle the challenge, we propose a task prefix guided multi-task pre-training framework to explore the relationships among tasks. We conduct extensive experiments on 40 datasets, which show that our model can not only serve as the strong foundation backbone for a wide range of tasks but also be feasible as a probing tool for analyzing task relationships. The task relationships reflected by the prefixes align transfer learning performance between tasks. They also suggest directions for data augmentation with complementary tasks, which help our model achieve human-parity results on commonsense reasoning leaderboards. Code is available at https://github.com/cooelf/CompassMTL

Non-autoregressive automatic speech recognition (ASR) modeling has received increasing attention recently because of its fast decoding speed and superior performance. Among representatives, methods based on the connectionist temporal classification (CTC) are still a dominating stream. However, the theoretically inherent flaw, the assumption of independence between tokens, creates a performance barrier for the school of works. To mitigate the challenge, we propose a context-aware knowledge transferring strategy, consisting of a knowledge transferring module and a context-aware training strategy, for CTC-based ASR. The former is designed to distill linguistic information from a pre-trained language model, and the latter is framed to modulate the limitations caused by the conditional independence assumption. As a result, a knowledge-injected context-aware CTC-based ASR built upon the wav2vec2.0 is presented in this paper. A series of experiments on the AISHELL-1 and AISHELL-2 datasets demonstrate the effectiveness of the proposed method.

Estimating the effects of continuous-valued interventions from observational data is a critically important task for climate science, healthcare, and economics. Recent work focuses on designing neural network architectures and regularization functions to allow for scalable estimation of average and individual-level dose-response curves from high-dimensional, large-sample data. Such methodologies assume ignorability (observation of all confounding variables) and positivity (observation of all treatment levels for every covariate value describing a set of units), assumptions problematic in the continuous treatment regime. Scalable sensitivity and uncertainty analyses to understand the ignorance induced in causal estimates when these assumptions are relaxed are less studied. Here, we develop a continuous treatment-effect marginal sensitivity model (CMSM) and derive bounds that agree with the observed data and a researcher-defined level of hidden confounding. We introduce a scalable algorithm and uncertainty-aware deep models to derive and estimate these bounds for high-dimensional, large-sample observational data. We work in concert with climate scientists interested in the climatological impacts of human emissions on cloud properties using satellite observations from the past 15 years. This problem is known to be complicated by many unobserved confounders.

Learning from non-stationary data streams, also called Task-Free Continual Learning (TFCL) remains challenging due to the absence of explicit task information. Although recently some methods have been proposed for TFCL, they lack theoretical guarantees. Moreover, forgetting analysis during TFCL was not studied theoretically before. This paper develops a new theoretical analysis framework which provides generalization bounds based on the discrepancy distance between the visited samples and the entire information made available for training the model. This analysis gives new insights into the forgetting behaviour in classification tasks. Inspired by this theoretical model, we propose a new approach enabled by the dynamic component expansion mechanism for a mixture model, namely the Online Discrepancy Distance Learning (ODDL). ODDL estimates the discrepancy between the probabilistic representation of the current memory buffer and the already accumulated knowledge and uses it as the expansion signal to ensure a compact network architecture with optimal performance. We then propose a new sample selection approach that selectively stores the most relevant samples into the memory buffer through the discrepancy-based measure, further improving the performance. We perform several TFCL experiments with the proposed methodology, which demonstrate that the proposed approach achieves the state of the art performance.