Research papers and code for "Zhifeng Chen":
Network slicing promises to provision diversified services with distinct requirements in one infrastructure. Deep reinforcement learning (e.g., deep $\mathcal{Q}$-learning, DQL) is assumed to be an appropriate algorithm to solve the demand-aware inter-slice resource management issue in network slicing by regarding the varying demands and the allocated bandwidth as the environment state and the action, respectively. However, allocating bandwidth in a finer resolution usually implies larger action space, and unfortunately DQL fails to quickly converge in this case. In this paper, we introduce discrete normalized advantage functions (DNAF) into DQL, by separating the $\mathcal{Q}$-value function as a state-value function term and an advantage term and exploiting a deterministic policy gradient descent (DPGD) algorithm to avoid the unnecessary calculation of $\mathcal{Q}$-value for every state-action pair. Furthermore, as DPGD only works in continuous action space, we embed a k-nearest neighbor algorithm into DQL to quickly find a valid action in the discrete space nearest to the DPGD output. Finally, we verify the faster convergence of the DNAF-based DQL through extensive simulations.

Click to Read Paper and Get Code
Network slicing is a key technology in 5G communications system, which aims to dynamically and efficiently allocate resources for diversified services with distinct requirements over a common underlying physical infrastructure. Therein, demand-aware allocation is of significant importance to network slicing. In this paper, we consider a scenario that contains several slices in one base station on sharing the same bandwidth. Deep reinforcement learning (DRL) is leveraged to solve this problem by regarding the varying demands and the allocated bandwidth as the environment \emph{state} and \emph{action}, respectively. In order to obtain better quality of experience (QoE) satisfaction ratio and spectrum efficiency (SE), we propose generative adversarial network (GAN) based deep distributional Q network (GAN-DDQN) to learn the distribution of state-action values. Furthermore, we estimate the distributions by approximating a full quantile function, which can make the training error more controllable. In order to protect the stability of GAN-DDQN's training process from the widely-spanning utility values, we also put forward a reward-clipping mechanism. Finally, we verify the performance of the proposed GAN-DDQN algorithm through extensive simulations.

* 7 pages, 5 figures, 22 conferences
Click to Read Paper and Get Code
Recent works have validated the possibility of improving energy efficiency in radio access networks (RANs), achieved by dynamically turning on/off some base stations (BSs). In this paper, we extend the research over BS switching operations, which should match up with traffic load variations. Instead of depending on the dynamic traffic loads which are still quite challenging to precisely forecast, we firstly formulate the traffic variations as a Markov decision process. Afterwards, in order to foresightedly minimize the energy consumption of RANs, we design a reinforcement learning framework based BS switching operation scheme. Furthermore, to avoid the underlying curse of dimensionality in reinforcement learning, a transfer actor-critic algorithm (TACT), which utilizes the transferred learning expertise in historical periods or neighboring regions, is proposed and provably converges. In the end, we evaluate our proposed scheme by extensive simulations under various practical configurations and show that the proposed TACT algorithm contributes to a performance jumpstart and demonstrates the feasibility of significant energy efficiency improvement at the expense of tolerable delay performance.

* 11 figures, 30 pages, accepted in IEEE Transactions on Wireless Communications 2014. IEEE Trans. Wireless Commun., Feb. 2014
Click to Read Paper and Get Code
We present a recurrent encoder-decoder deep neural network architecture that directly translates speech in one language into text in another. The model does not explicitly transcribe the speech into text in the source language, nor does it require supervision from the ground truth source language transcription during training. We apply a slightly modified sequence-to-sequence with attention architecture that has previously been used for speech recognition and show that it can be repurposed for this more complex task, illustrating the power of attention-based models. A single model trained end-to-end obtains state-of-the-art performance on the Fisher Callhome Spanish-English speech translation task, outperforming a cascade of independently trained sequence-to-sequence speech recognition and machine translation models by 1.8 BLEU points on the Fisher test set. In addition, we find that making use of the training data in both languages by multi-task training sequence-to-sequence speech translation and recognition models with a shared encoder network can improve performance by a further 1.4 BLEU points.

* 5 pages, 1 figure. Interspeech 2017
Click to Read Paper and Get Code
Time series prediction can be generalized as a process that extracts useful information from historical records and then determines future values. Learning long-range dependencies that are embedded in time series is often an obstacle for most algorithms, whereas Long Short-Term Memory (LSTM) solutions, as a specific kind of scheme in deep learning, promise to effectively overcome the problem. In this article, we first give a brief introduction to the structure and forward propagation mechanism of the LSTM model. Then, aiming at reducing the considerable computing cost of LSTM, we put forward the Random Connectivity LSTM (RCLSTM) model and test it by predicting traffic and user mobility in telecommunication networks. Compared to LSTM, RCLSTM is formed via stochastic connectivity between neurons, which achieves a significant breakthrough in the architecture formation of neural networks. In this way, the RCLSTM model exhibits a certain level of sparsity, which leads to an appealing decrease in the computational complexity and makes the RCLSTM model become more applicable in latency-stringent application scenarios. In the field of telecommunication networks, the prediction of traffic series and mobility traces could directly benefit from this improvement as we further demonstrate that the prediction accuracy of RCLSTM is comparable to that of the conventional LSTM no matter how we change the number of training samples or the length of input sequences.

* 9 pages, 5 figures, 14 references
Click to Read Paper and Get Code
Traffic prediction plays an important role in evaluating the performance of telecommunication networks and attracts intense research interests. A significant number of algorithms and models have been put forward to analyse traffic data and make prediction. In the recent big data era, deep learning has been exploited to mine the profound information hidden in the data. In particular, Long Short-Term Memory (LSTM), one kind of Recurrent Neural Network (RNN) schemes, has attracted a lot of attentions due to its capability of processing the long-range dependency embedded in the sequential traffic data. However, LSTM has considerable computational cost, which can not be tolerated in tasks with stringent latency requirement. In this paper, we propose a deep learning model based on LSTM, called Random Connectivity LSTM (RCLSTM). Compared to the conventional LSTM, RCLSTM makes a notable breakthrough in the formation of neural network, which is that the neurons are connected in a stochastic manner rather than full connected. So, the RCLSTM, with certain intrinsic sparsity, have many neural connections absent (distinguished from the full connectivity) and which leads to the reduction of the parameters to be trained and the computational cost. We apply the RCLSTM to predict traffic and validate that the RCLSTM with even 35% neural connectivity still shows a satisfactory performance. When we gradually add training samples, the performance of RCLSTM becomes increasingly closer to the baseline LSTM. Moreover, for the input traffic sequences of enough length, the RCLSTM exhibits even superior prediction accuracy than the baseline LSTM.

* 6 pages, 9 figures
Click to Read Paper and Get Code
Attention-based sequence-to-sequence models for automatic speech recognition jointly train an acoustic model, language model, and alignment mechanism. Thus, the language model component is only trained on transcribed audio-text pairs. This leads to the use of shallow fusion with an external language model at inference time. Shallow fusion refers to log-linear interpolation with a separately trained language model at each step of the beam search. In this work, we investigate the behavior of shallow fusion across a range of conditions: different types of language models, different decoding units, and different tasks. On Google Voice Search, we demonstrate that the use of shallow fusion with a neural LM with wordpieces yields a 9.1% relative word error rate reduction (WERR) over our competitive attention-based sequence-to-sequence model, obviating the need for second-pass rescoring.

Click to Read Paper and Get Code
Many real-world applications reveal difficulties in learning classifiers from imbalanced data. The rising big data era has been witnessing more classification tasks with large-scale but extremely imbalance and low-quality datasets. Most of existing learning methods suffer from poor performance or low computation efficiency under such a scenario. To tackle this problem, we conduct deep investigations into the nature of class imbalance, which reveals that not only the disproportion between classes, but also other difficulties embedded in the nature of data, especially, noises and class overlapping, prevent us from learning effective classifiers. Taking those factors into consideration, we propose a novel framework for imbalance classification that aims to generate a strong ensemble by self-paced harmonizing data hardness via under-sampling. Extensive experiments have shown that this new framework, while being very computationally efficient, can lead to robust performance even under highly overlapping classes and extremely skewed distribution. Note that, our methods can be easily adapted to most of existing learning methods (e.g., C4.5, SVM, GBDT and Neural Network) to boost their performance on imbalanced data.

Click to Read Paper and Get Code
GPipe is a scalable pipeline parallelism library that enables learning of giant deep neural networks. It partitions network layers across accelerators and pipelines execution to achieve high hardware utilization. It leverages recomputation to minimize activation memory usage. For example, using partitions over 8 accelerators, it is able to train networks that are 25x larger, demonstrating its scalability. It also guarantees that the computed gradients remain consistent regardless of the number of partitions. It achieves an almost linear speed up without any changes in the model parameters: when using 4x more accelerators, training the same model is up to 3.5x faster. We train a 557 million parameters AmoebaNet model on ImageNet and achieve a new state-of-the-art 84.3% top-1 / 97.0% top-5 accuracy on ImageNet 2012 dataset. Finally, we use this learned model to finetune multiple popular image classification datasets and obtain competitive results, including pushing the CIFAR-10 accuracy to 99% and CIFAR-100 accuracy to 91.3%.

* 10 pages. Work in progress. Copyright 2018 by the authors
Click to Read Paper and Get Code
A key problem in structured output prediction is direct optimization of the task reward function that matters for test evaluation. This paper presents a simple and computationally efficient approach to incorporate task reward into a maximum likelihood framework. By establishing a link between the log-likelihood and expected reward objectives, we show that an optimal regularized expected reward is achieved when the conditional distribution of the outputs given the inputs is proportional to their exponentiated scaled rewards. Accordingly, we present a framework to smooth the predictive probability of the outputs using their corresponding rewards. We optimize the conditional log-probability of augmented outputs that are sampled proportionally to their exponentiated scaled rewards. Experiments on neural sequence to sequence models for speech recognition and machine translation show notable improvements over a maximum likelihood baseline by using reward augmented maximum likelihood (RAML), where the rewards are defined as the negative edit distance between the outputs and the ground truth labels.

* NIPS 2016
Click to Read Paper and Get Code
We present an attention-based sequence-to-sequence neural network which can directly translate speech from one language into speech in another language, without relying on an intermediate text representation. The network is trained end-to-end, learning to map speech spectrograms into target spectrograms in another language, corresponding to the translated content (in a different canonical voice). We further demonstrate the ability to synthesize translated speech using the voice of the source speaker. We conduct experiments on two Spanish-to-English speech translation datasets, and find that the proposed model slightly underperforms a baseline cascade of a direct speech-to-text translation model and a text-to-speech synthesis model, demonstrating the feasibility of the approach on this very challenging task.

* Submitted to Interspeech 2019
Click to Read Paper and Get Code
Network slicing means an emerging business to operators and allows them to sell the customized slices to various tenants at different prices. In order to provide better-performing and costefficient services, network slicing involves challenging technical issues and urgently looks forward to intelligent innovations to make the resource management consistent with users' activities per slice. In that regard, deep reinforcement learning (DRL), which focuses on how to interact with the environment by trying alternative actions and reinforces the tendency actions producing more rewarding consequences, is emerging as a promising solution. In this paper, after briefly reviewing the fundamental concepts and evolution-driving factors of DRL, we investigate the application of DRL in some typical resource management scenarios of network slicing, which include radio resource slicing and priority-based core network slicing, and demonstrate the performance advantage of DRL over several competing schemes through extensive simulations. Finally, we also discuss the possible challenges to apply DRL in network slicing from a general perspective.

* 5 figures, 1 table
Click to Read Paper and Get Code
Sequence-to-sequence models, such as attention-based models in automatic speech recognition (ASR), are typically trained to optimize the cross-entropy criterion which corresponds to improving the log-likelihood of the data. However, system performance is usually measured in terms of word error rate (WER), not log-likelihood. Traditional ASR systems benefit from discriminative sequence training which optimizes criteria such as the state-level minimum Bayes risk (sMBR) which are more closely related to WER. In the present work, we explore techniques to train attention-based models to directly minimize expected word error rate. We consider two loss functions which approximate the expected number of word errors: either by sampling from the model, or by using N-best lists of decoded hypotheses, which we find to be more effective than the sampling-based method. In experimental evaluations, we find that the proposed training procedure improves performance by up to 8.2% relative to the baseline system. This allows us to train grapheme-based, uni-directional attention-based models which match the performance of a traditional, state-of-the-art, discriminative sequence-trained system on a mobile voice-search task.

Click to Read Paper and Get Code
Having a sequence-to-sequence model which can operate in an online fashion is important for streaming applications such as Voice Search. Neural transducer is a streaming sequence-to-sequence model, but has shown a significant degradation in performance compared to non-streaming models such as Listen, Attend and Spell (LAS). In this paper, we present various improvements to NT. Specifically, we look at increasing the window over which NT computes attention, mainly by looking backwards in time so the model still remains online. In addition, we explore initializing a NT model from a LAS-trained model so that it is guided with a better alignment. Finally, we explore including stronger language models such as using wordpiece models, and applying an external LM during the beam search. On a Voice Search task, we find with these improvements we can get NT to match the performance of LAS.

Click to Read Paper and Get Code
Multilingual end-to-end (E2E) models have shown great promise in expansion of automatic speech recognition (ASR) coverage of the world's languages. They have shown improvement over monolingual systems, and have simplified training and serving by eliminating language-specific acoustic, pronunciation, and language models. This work presents an E2E multilingual system which is equipped to operate in low-latency interactive applications, as well as handle a key challenge of real world data: the imbalance in training data across languages. Using nine Indic languages, we compare a variety of techniques, and find that a combination of conditioning on a language vector and training language-specific adapter layers produces the best model. The resulting E2E multilingual model achieves a lower word error rate (WER) than both monolingual E2E models (eight of nine languages) and monolingual conventional systems (all nine languages).

* Accepted in Interspeech 2019
Click to Read Paper and Get Code
We present a multispeaker, multilingual text-to-speech (TTS) synthesis model based on Tacotron that is able to produce high quality speech in multiple languages. Moreover, the model is able to transfer voices across languages, e.g. synthesize fluent Spanish speech using an English speaker's voice, without training on any bilingual or parallel examples. Such transfer works across distantly related languages, e.g. English and Mandarin. Critical to achieving this result are: 1. using a phonemic input representation to encourage sharing of model capacity across languages, and 2. incorporating an adversarial loss term to encourage the model to disentangle its representation of speaker identity (which is perfectly correlated with language in the training data) from the speech content. Further scaling up the model by training on multiple speakers of each language, and incorporating an autoencoding input to help stabilize attention during training, results in a model which can be used to consistently synthesize intelligible speech for training speakers in all languages seen during training, and in native or foreign accents.

* 5 pages, submitted to Interspeech 2019
Click to Read Paper and Get Code
We describe a neural network-based system for text-to-speech (TTS) synthesis that is able to generate speech audio in the voice of many different speakers, including those unseen during training. Our system consists of three independently trained components: (1) a speaker encoder network, trained on a speaker verification task using an independent dataset of noisy speech from thousands of speakers without transcripts, to generate a fixed-dimensional embedding vector from seconds of reference speech from a target speaker; (2) a sequence-to-sequence synthesis network based on Tacotron 2, which generates a mel spectrogram from text, conditioned on the speaker embedding; (3) an auto-regressive WaveNet-based vocoder that converts the mel spectrogram into a sequence of time domain waveform samples. We demonstrate that the proposed model is able to transfer the knowledge of speaker variability learned by the discriminatively-trained speaker encoder to the new task, and is able to synthesize natural speech from speakers that were not seen during training. We quantify the importance of training the speaker encoder on a large and diverse speaker set in order to obtain the best generalization performance. Finally, we show that randomly sampled speaker embeddings can be used to synthesize speech in the voice of novel speakers dissimilar from those used in training, indicating that the model has learned a high quality speaker representation.

* NIPS 2018
Click to Read Paper and Get Code
The past year has witnessed rapid advances in sequence-to-sequence (seq2seq) modeling for Machine Translation (MT). The classic RNN-based approaches to MT were first out-performed by the convolutional seq2seq model, which was then out-performed by the more recent Transformer model. Each of these new approaches consists of a fundamental architecture accompanied by a set of modeling and training techniques that are in principle applicable to other seq2seq architectures. In this paper, we tease apart the new architectures and their accompanying techniques in two ways. First, we identify several key modeling and training techniques, and apply them to the RNN architecture, yielding a new RNMT+ model that outperforms all of the three fundamental architectures on the benchmark WMT'14 English to French and English to German tasks. Second, we analyze the properties of each fundamental seq2seq architecture and devise new hybrid architectures intended to combine their strengths. Our hybrid models obtain further improvements, outperforming the RNMT+ model on both benchmark datasets.

Click to Read Paper and Get Code
We propose a simple solution to use a single Neural Machine Translation (NMT) model to translate between multiple languages. Our solution requires no change in the model architecture from our base system but instead introduces an artificial token at the beginning of the input sentence to specify the required target language. The rest of the model, which includes encoder, decoder and attention, remains unchanged and is shared across all languages. Using a shared wordpiece vocabulary, our approach enables Multilingual NMT using a single model without any increase in parameters, which is significantly simpler than previous proposals for Multilingual NMT. Our method often improves the translation quality of all involved language pairs, even while keeping the total number of model parameters constant. On the WMT'14 benchmarks, a single multilingual model achieves comparable performance for English$\rightarrow$French and surpasses state-of-the-art results for English$\rightarrow$German. Similarly, a single multilingual model surpasses state-of-the-art results for French$\rightarrow$English and German$\rightarrow$English on WMT'14 and WMT'15 benchmarks respectively. On production corpora, multilingual models of up to twelve language pairs allow for better translation of many individual pairs. In addition to improving the translation quality of language pairs that the model was trained with, our models can also learn to perform implicit bridging between language pairs never seen explicitly during training, showing that transfer learning and zero-shot translation is possible for neural translation. Finally, we show analyses that hints at a universal interlingua representation in our models and show some interesting examples when mixing languages.

Click to Read Paper and Get Code