Robert Gmyr

Ziyi Yang, Yuwei Fang, Chenguang Zhu et al. · gatech, stanford

Human intelligence is multimodal; we integrate visual, linguistic, and acoustic signals to maintain a holistic worldview. Most current pretraining methods, however, are limited to one or two modalities. We present i-Code, a self-supervised pretraining framework where users may flexibly combine the modalities of vision, speech, and language into unified and general-purpose vector representations. In this framework, data from each modality are first given to pretrained single-modality encoders. The encoder outputs are then integrated with a multimodal fusion network, which uses novel attention mechanisms and other architectural innovations to effectively combine information from the different modalities. The entire system is pretrained end-to-end with new objectives including masked modality unit modeling and cross-modality contrastive learning. Unlike previous research using only video for pretraining, the i-Code framework can dynamically process single, dual, and triple-modality data during training and inference, flexibly projecting different combinations of modalities into a single representation space. Experimental results demonstrate how i-Code can outperform state-of-the-art techniques on five video understanding tasks and the GLUE NLP benchmark, improving by as much as 11% and demonstrating the power of integrative multimodal pretraining.

Ziyi Yang, Mahmoud Khademi, Yichong Xu et al.

The convergence of text, visual, and audio data is a key step towards human-like artificial intelligence, however the current Vision-Language-Speech landscape is dominated by encoder-only models which lack generative abilities. We propose closing this gap with i-Code V2, the first model capable of generating natural language from any combination of Vision, Language, and Speech data. i-Code V2 is an integrative system that leverages state-of-the-art single-modality encoders, combining their outputs with a new modality-fusing encoder in order to flexibly project combinations of modalities into a shared representational space. Next, language tokens are generated from these representations via an autoregressive decoder. The whole framework is pretrained end-to-end on a large collection of dual- and single-modality datasets using a novel text completion objective that can be generalized across arbitrary combinations of modalities. i-Code V2 matches or outperforms state-of-the-art single- and dual-modality baselines on 7 multimodal tasks, demonstrating the power of generative multimodal pretraining across a diversity of tasks and signals.

Kenichi Kumatani, Dimitrios Dimitriadis, Yashesh Gaur et al.

In this work, we develop new self-learning techniques with an attention-based sequence-to-sequence (seq2seq) model for automatic speech recognition (ASR). For untranscribed speech data, the hypothesis from an ASR system must be used as a label. However, the imperfect ASR result makes unsupervised learning difficult to consistently improve recognition performance especially in the case that multiple powerful teacher models are unavailable. In contrast to conventional unsupervised learning approaches, we adopt the \emph{multi-task learning} (MTL) framework where the $n$-th best ASR hypothesis is used as the label of each task. The seq2seq network is updated through the MTL framework so as to find the common representation that can cover multiple hypotheses. By doing so, the effect of the \emph{hard-decision} errors can be alleviated. We first demonstrate the effectiveness of our self-learning methods through ASR experiments in an accent adaptation task between the US and British English speech. Our experiment results show that our method can reduce the WER on the British speech data from 14.55\% to 10.36\% compared to the baseline model trained with the US English data only. Moreover, we investigate the effect of our proposed methods in a federated learning scenario.

Kenichi Kumatani, Robert Gmyr, Felipe Cruz Salinas et al.

The sparsely-gated Mixture of Experts (MoE) can magnify a network capacity with a little computational complexity. In this work, we investigate how multi-lingual Automatic Speech Recognition (ASR) networks can be scaled up with a simple routing algorithm in order to achieve better accuracy. More specifically, we apply the sparsely-gated MoE technique to two types of networks: Sequence-to-Sequence Transformer (S2S-T) and Transformer Transducer (T-T). We demonstrate through a set of ASR experiments on multiple language data that the MoE networks can reduce the relative word error rates by 16.3% and 4.6% with the S2S-T and T-T, respectively. Moreover, we thoroughly investigate the effect of the MoE on the T-T architecture in various conditions: streaming mode, non-streaming mode, the use of language ID and the label decoder with the MoE.

Dimitrios Dimitriadis, Kenichi Kumatani, Robert Gmyr et al.

In this paper, a new learning algorithm for Federated Learning (FL) is introduced. The proposed scheme is based on a weighted gradient aggregation using two-step optimization to offer a flexible training pipeline. Herein, two different flavors of the aggregation method are presented, leading to an order of magnitude improvement in convergence speed compared to other distributed or FL training algorithms like BMUF and FedAvg. Further, the aggregation algorithm acts as a regularizer of the gradient quality. We investigate the effect of our FL algorithm in supervised and unsupervised Speech Recognition (SR) scenarios. The experimental validation is performed based on three tasks: first, the LibriSpeech task showing a speed-up of 7x and 6% word error rate reduction (WERR) compared to the baseline results. The second task is based on session adaptation providing 20% WERR over a powerful LAS model. Finally, our unsupervised pipeline is applied to the conversational SR task. The proposed FL system outperforms the baseline systems in both convergence speed and overall model performance.

Dimitrios Dimitriadis, Kenichi Kumatani, Robert Gmyr et al.

In this paper, a Federated Learning (FL) simulation platform is introduced. The target scenario is Acoustic Model training based on this platform. To our knowledge, this is the first attempt to apply FL techniques to Speech Recognition tasks due to the inherent complexity. The proposed FL platform can support different tasks based on the adopted modular design. As part of the platform, a novel hierarchical optimization scheme and two gradient aggregation methods are proposed, leading to almost an order of magnitude improvement in training convergence speed compared to other distributed or FL training algorithms like BMUF and FedAvg. The hierarchical optimization offers additional flexibility in the training pipeline besides the enhanced convergence speed. On top of the hierarchical optimization, a dynamic gradient aggregation algorithm is proposed, based on a data-driven weight inference. This aggregation algorithm acts as a regularizer of the gradient quality. Finally, an unsupervised training pipeline tailored to FL is presented as a separate training scenario. The experimental validation of the proposed system is based on two tasks: first, the LibriSpeech task showing a speed-up of 7x and 6% Word Error Rate reduction (WERR) compared to the baseline results. The second task is based on session adaptation providing an improvement of 20% WERR over a competitive production-ready LAS model. The proposed Federated Learning system is shown to outperform the golden standard of distributed training in both convergence speed and overall model performance.

Ziyi Yang, Chenguang Zhu, Robert Gmyr et al.

Text summarization aims to extract essential information from a piece of text and transform the text into a concise version. Existing unsupervised abstractive summarization models leverage recurrent neural networks framework while the recently proposed transformer exhibits much more capability. Moreover, most of previous summarization models ignore abundant unlabeled corpora resources available for pretraining. In order to address these issues, we propose TED, a transformer-based unsupervised abstractive summarization system with pretraining on large-scale data. We first leverage the lead bias in news articles to pretrain the model on millions of unlabeled corpora. Next, we finetune TED on target domains through theme modeling and a denoising autoencoder to enhance the quality of generated summaries. Notably, TED outperforms all unsupervised abstractive baselines on NYT, CNN/DM and English Gigaword datasets with various document styles. Further analysis shows that the summaries generated by TED are highly abstractive, and each component in the objective function of TED is highly effective.

Chenguang Zhu, Ziyi Yang, Robert Gmyr et al.

A typical journalistic convention in news articles is to deliver the most salient information in the beginning, also known as the lead bias. While this phenomenon can be exploited in generating a summary, it has a detrimental effect on teaching a model to discriminate and extract important information in general. We propose that this lead bias can be leveraged in our favor in a simple and effective way to pre-train abstractive news summarization models on large-scale unlabeled news corpora: predicting the leading sentences using the rest of an article. We collect a massive news corpus and conduct data cleaning and filtering via statistical analysis. We then apply self-supervised pre-training on this dataset to existing generation models BART and T5 for domain adaptation. Via extensive experiments on six benchmark datasets, we show that this approach can dramatically improve the summarization quality and achieve state-of-the-art results for zero-shot news summarization without any fine-tuning. For example, in the DUC2003 dataset, the ROUGE-1 score of BART increases 13.7% after the lead-bias pre-training. We deploy the model in Microsoft News and provide public APIs as well as a demo website for multi-lingual news summarization.

Joshua J. Daymude, Robert Gmyr, Kristian Hinnenthal et al.

We envision programmable matter as a system of nano-scale agents (called particles) with very limited computational capabilities that move and compute collectively to achieve a desired goal. We use the geometric amoebot model as our computational framework, which assumes particles move on the triangular lattice. Motivated by the problem of sealing an object using minimal resources, we show how a particle system can self-organize to form an object's convex hull. We give a distributed, local algorithm for convex hull formation and prove that it runs in $\mathcal{O}(B)$ asynchronous rounds, where $B$ is the length of the object's boundary. Within the same asymptotic runtime, this algorithm can be extended to also form the object's (weak) $\mathcal{O}$-hull, which uses the same number of particles but minimizes the area enclosed by the hull. Our algorithms are the first to compute convex hulls with distributed entities that have strictly local sensing, constant-size memory, and no shared sense of orientation or coordinates. Ours is also the first distributed approach to computing restricted-orientation convex hulls. This approach involves coordinating particles as distributed memory; thus, as a supporting but independent result, we present and analyze an algorithm for organizing particles with constant-size memory as distributed binary counters that efficiently support increments, decrements, and zero-tests --- even as the particles move.

Joshua J. Daymude, Robert Gmyr, Andrea W. Richa et al.

We consider programmable matter that consists of computationally limited devices (called particles) that are able to self-organize in order to achieve some collective goal without the need for central control or external intervention. We use the geometric amoebot model to describe such self-organizing particle systems, which defines how particles can actively move and communicate with one another. In this paper, we present an efficient local-control algorithm which solves the leader election problem in O(n) asynchronous rounds with high probability, where n is the number of particles in the system. Our algorithm relies only on local information --- particles do not have unique identifiers, any knowledge of n, or any sort of global coordinate system --- and requires only constant memory per particle.