# stcn_stochastic_temporal_convolutional_networks__3d255c13.pdf

Published as a conference paper at ICLR 2019

STCN: STOCHASTIC TEMPORAL CONVOLUTIONAL NETWORKS

Emre Aksan & Otmar Hilliges Department of Computer Science ETH Zurich, Switzerland {emre.aksan, otmar.hilliges}@inf.ethz.ch

Convolutional architectures have recently been shown to be competitive on many sequence modelling tasks when compared to the de-facto standard of recurrent neural networks (RNNs), while providing computational and modeling advantages due to inherent parallelism. However, currently there remains a performance gap to more expressive stochastic RNN variants, especially those with several layers of dependent random variables. In this work, we propose stochastic temporal convolutional networks (STCNs), a novel architecture that combines the computational advantages of temporal convolutional networks (TCN) with the representational power and robustness of stochastic latent spaces. In particular, we propose a hierarchy of stochastic latent variables that captures temporal dependencies at different time-scales. The architecture is modular and flexible due to decoupling of deterministic and stochastic layers. We show that the proposed architecture achieves state of the art log-likelihoods across several tasks. Finally, the model is capable of predicting high-quality synthetic samples over a long-range temporal horizon in modeling of handwritten text.

1 INTRODUCTION

Generative modeling of sequence data requires capturing long-term dependencies and learning of correlations between output variables at the same time-step. Recurrent neural networks (RNNs) and its variants have been very successful in a vast number of problem domains which rely on sequential data. Recent work in audio synthesis, language modeling and machine translation tasks (Dauphin et al., 2016; Van Den Oord et al., 2016; Dieleman et al., 2018; Gehring et al., 2017) has demonstrated that temporal convolutional networks (TCNs) can also achieve at least competitive performance without relying on recurrence, and hence reducing the computational cost for training.

Both RNNs and TCNs model the joint probability distribution over sequences by decomposing the distribution over discrete time-steps. In other words, such models are trained to predict the next step, given all previous time-steps. RNNs are able to model long-term dependencies by propagating information through their deterministic hidden state, acting as an internal memory. In contrast, TCNs leverage large receptive fields by stacking many dilated convolutions, allowing them to model even longer time scales up to the entire sequence length. It is noteworthy that there is no explicit temporal dependency between the model outputs and hence the computations can be performed in parallel. The TCN architecture also introduces a temporal hierarchy: the upper layers have access to longer input sub-sequences and learn representations at a larger time scale. The local information from the lower layers is propagated through the hierarchy by means of residual and skip connections (Van Den Oord et al., 2016; Bai et al., 2018).

However, while TCN architectures have been shown to perform similar or better than standard recurrent architectures on particular tasks (Van Den Oord et al., 2016; Bai et al., 2018), there currently remains a performance gap to more recent stochastic RNN variants (Bayer & Osendorfer, 2014; Chung et al., 2015; Fabius & van Amersfoort, 2014; Fraccaro et al., 2016; Goyal et al., 2017; Shabanian et al., 2017). Following a similar approach to stochastic RNNs, Lai et al. (2018) present a significant improvement in the log-likelihood when a TCN model is coupled with latent variables, albeit at the cost of limited receptive field size.

Published as a conference paper at ICLR 2019

Figure 1: The computational graph of generative (left) and inference (right) models of STCN. The approximate posterior q is conditioned on dt and is updated by the prior p which is conditioned on the TCN representations of the previous time-step dt 1. The random latent variables at the upper layers have access to a long history while lower layers receive inputs from more recent time steps.

In this work we propose a new approach for augmenting TCNs with random latent variables, that decouples deterministic and stochastic structures yet leverages the increased modeling capacity efficiently. Motivated by the simplicity and computational advantages of TCNs and the robustness and performance of stochastic RNNs, we introduce stochastic temporal convolutional networks (STCN) by incorporating a hierarchy of stochastic latent variables into TCNs which enables learning of representations at many timescales. However, due to the absence of an internal state in TCNs, introducing latent random variables analogously to stochastic RNNs is not feasible. Furthermore, defining conditional random variables across time-steps would result in breaking the parallelism of TCNs and is hence undesirable.

In STCN the latent random variables are arranged in correspondence to the temporal hierarchy of the TCN blocks, effectively distributing them over the various timescales (see figure 1). Crucially, our hierarchical latent structure is designed to be a modular add-on for any temporal convolutional network architecture. Separating the deterministic and stochastic layers allows us to build STCNs without requiring modifications to the base TCN architecture, and hence retains the scalability of TCNs with respect to the receptive field. This conditioning of the latent random variables via different timescales is especially effective in the case of TCNs. We show this experimentally by replacing the TCN layers with stacked LSTM cells, leading to reduced performance compared to STCN.

We propose two different inference networks. In the canonical configuration, samples from each latent variable are passed down from layer to layer and only one sample from the lowest layer is used to condition the prediction of the output. In the second configuration, called STCN-dense, we take inspiration from recent CNN architectures (Huang et al., 2017) and utilize samples from all latent random variables via concatenation before computing the final prediction.

Our contributions can thus be summarized as: 1) We present a modular and scalable approach to augment temporal convolutional network models with effective stochastic latent variables. 2) We empirically show that the STCN-dense design prevents the model from ignoring latent variables in the upper layers (Zhao et al., 2017). 3) We achieve state-of-the-art log-likelihood performance, measured by ELBO, on the IAM-On DB, Deepwriting, TIMIT and the Blizzard datasets. 4) Finally we show that the quality of the synthetic samples matches the significant quantitative improvements.

2 BACKGROUND

Auto-regressive models such as RNNs and TCNs factorize the joint probability of a variable-length sequence x = {x1, . . . , x T } as a product of conditionals as follows:

t=1 pθ(xt|x1:t 1) , (1)

where the joint distribution is parametrized by θ. The prediction at each time-step is conditioned on all previous observations. The observation model is frequently chosen to be a Gaussian or Gaussian mixture model (GMM) for real-valued data, and a categorical distribution for discrete-valued data.

Published as a conference paper at ICLR 2019

2.1 TEMPORAL CONVOLUTIONAL NETWORKS

In TCNs the joint probabilities in Eq. (1) are parametrized by a stack of convolutional layers. Causal convolutions are the central building block of such models and are designed to be asymmetric such that the model has no access to future information. In order to produce outputs of the same size as the input, zero-padding is applied at every layer.

In the absence of a state transition function, a large receptive field is crucial in capturing long-range dependencies. To avoid the need for vast numbers of causal convolution layers, typically dilated convolutions are used. Exponentially increasing the dilation factor results in an exponential growth of the receptive field size with depth (Yu & Koltun, 2015; Van Den Oord et al., 2016; Bai et al., 2018). In this work, without loss of generality, we use the building blocks of Wavenet (Van Den Oord et al., 2016) as gated activation units (van den Oord et al., 2016) have been reported to perform better.

A deterministic TCN representation dl t at time-step t and layer l summarizes the input sequence x1:t:

dl t = Conv(l)(dl 1 t , dl 1 t j) and d1 t = Conv(1)(xt, xt j) , (2)

where the filter width is 2 and j denotes the dilation step. In our work, the stochastic variables zl, l = 1 . . . L are conditioned on TCN representations dl that are constructed by stacking K Wavenet blocks over the previous dl 1 (for details see Figure 4 in Appendix).

2.2 NON-SEQUENTIAL LATENT VARIABLE MODELS

VAEs (Kingma & Welling, 2013; Rezende et al., 2014) introduce a latent random variable z to learn the variations in the observed non-sequential data where the generation of the sample x is conditioned on the latent variable z. The joint probability distribution is defined as:

pθ(x, z) = pθ(x|z)pθ(z) , (3)

and parametrized by θ. Optimizing the marginal likelihood is intractable due to the non-linear mappings between z and x and the integration over z. Instead the VAE framework introduces an approximate posterior qφ(z|x) and optimizes a lower-bound on the marginal likelihood:

log pθ(x) KL(qφ(z|x)||pθ(z)) + Eqφ(z|x)[log pθ(x|z)] , (4)

where KL denotes the Kullback-Leibler divergence. Typically the prior pθ(z) and the approximate qφ(z|x) are chosen to be in simple parametric form, such as a Gaussian distribution with diagonal covariance, which allows for an analytical calculation of the KL-term in Eq. (4).

2.3 STOCHASTIC RNNS

An RNN captures temporal dependencies by recursively processing each input, while updating an internal state ht at each time-step via its state-transition function:

ht = f (h)(xt, ht 1) , (5)

where f (h) is a deterministic transition function such as LSTM (Hochreiter & Schmidhuber, 1997) or GRU (Cho et al., 2014) cells. The computation has to be sequential because ht depends on ht 1.

The VAE framework has been extended for sequential data, where a latent variable zt augments the RNN state ht at each sequence step. The joint distribution pθ(x, z) is modeled via an auto-regressive model which results in the following factorization:

t=1 pθ(xt|z1:t, x1:t 1)pθ(zt|x1:t 1, z1:t 1) . (6)

In contrast to the fixed prior of VAEs, N(0, I), sequential variants define prior distributions conditioned on the RNN hidden state h and implicitly on the input sequence x (Chung et al., 2015).

Published as a conference paper at ICLR 2019

Figure 2: Graphical model view of generative models of STCN (left) and STCN-dense (middle), and the inference model (right), which is shared by both variants. Diamonds represent the outputs of deterministic dilated convolution blocks where the dependence of dt on the past inputs is not shown for clarity (see Eq. (2)). xt and zt are observable inputs and latent random variables, respectively. The generative task is to predict the next step in the sequence, given all past steps. Note that in the STCN-dense variant the next step is conditioned on all latent variables zl t for l = 1 . . . L.

3 STOCHASTIC TEMPORAL CONVOLUTIONAL NETWORKS

The mechanics of STCNs are related to those of VRNNs and LVAEs. Intuitively, the RNN state ht is replaced by temporally independent TCN layers dl t. In the absence of an internal state, we define hierarchical latent variables zl t that are conditioned vertically, i.e., in the same time-step, but independent horizontally, i.e., across time-steps. We follow a similar approach to LVAEs (Sønderby et al., 2016) in defining the hierarchy in a top-down fashion and in how we estimate the approximate posterior. The inference network first computes the approximate likelihood, and then this estimate is corrected by the prior, resulting in the approximate posterior. The TCN layers d are shared between the inference and generator networks, analogous to VRNNs (Chung et al., 2015).

Figure 2 depicts the proposed STCN as a graphical model. STCNs consist of two main modules: the deterministic temporal convolutional network and the stochastic latent variable hierarchy. For a given input sequence x = {xt}, t = 1 . . . T we first apply dilated convolutions over the entire sequence to compute a set of deterministic representations dl t, l = 1 . . . L. Here, dl t corresponds to the output of a block of dilated convolutions at layer l and time-step t. The output dl t is then used to update a set of random latent variables zl t arranged to correspond with different time-scales.

To preserve the parallelism of TCNs, we do not introduce an explicit dependency between different time-steps. However, we suggest that conditioning a latent variable zl 1 t on the preceding variable zl t implicitly introduces temporal dependencies. Importantly, the random latent variables in the upper layer have access to a larger receptive field due to its deterministic input dl t 1, whereas latent random variables in lower layers are updated with different, more local information. However, the latent variable zl 1 t may receive longer-range information from zl t.

The generative and inference models are jointly trained by optimizing a step-wise variational lower bound on the log-likelihood (Kingma & Welling, 2013; Rezende et al., 2014). In the following sections we describe these components and build up the lower-bound for a single time-step t.

3.1 GENERATIVE MODEL

Each sequence step xt is generated from a set of latent variables zt, split into layers as follows:

pθ(zt|x1:t 1) = pθ(z L t |d L t 1)

l=1 pθ(zl t|zl+1 t , dl t 1) , (7)

where pθ(zl t|zl+1 t , dl t 1) = N(µl t,p, σl t,p) and [µl t,p, σl t,p] = f (l) p (zl+1 t , dl t 1) . (8)

Here the prior is modeled by a Gaussian distribution with diagonal covariance, as is common in the VAE framework. The subscript p denotes items of the generative distribution. For the inference distribution we use the subscript q. The distributions are parameterized by a neural network f (l) p and conditioned on: (1) the dl t 1 computed by the dilated convolutions from the previous time-step, and

Published as a conference paper at ICLR 2019

(2) a sample from the preceding level at the same time-step zl+1 t . Please note that at inference time we draw samples from the approximate posterior distribution zl+1 t qφ(zl+1 t | ). The generative model, on the other hand, uses the prior zl+1 t pθ(zl+1 t | ).

We propose two variants of the observation model. In the non-sequential scenario, the observations are defined to be conditioned on only the last latent variable in the hierarchy, i.e., pθ(xt|z1 t ), following Sønderby et al. (2016); Gulrajani et al. (2016) and Rezende et al. (2014) our STCN variant uses the same observation model, allowing for an efficient optimization. However, latent units are likely to become inactive during training in this configuration (Burda et al., 2015; Bowman et al., 2015; Zhao et al., 2017) resulting in a loss of representational power.

The latent variables at different layers are conditioned on different contexts due to the inputs dl t. Hence, the latent variables are expected to capture complementary aspects of the temporal context. To propagate the information all the way to the final prediction and to ensure that gradients flow through all layers, we take inspiration from Huang et al. (2017) and directly condition the output probability on samples from all latent variables. We call this variant of our architecture STCN-dense.

The final predictions are then computed by the respective observation functions:

pθ(xt|zt) = f (o)(z1 t ) and pdense θ (xt|zt) = f (o)(z1 t , z2 t . . . z L t ) , (9)

where f (o) corresponds to the output layer constructed by stacking 1D convolutions or Wavenet blocks depending on the dataset.

3.2 INFERENCE MODEL

In the original VAE framework the inference model is defined as a bottom-up process, where the latent variables are conditioned on the stochastic layer below. Furthermore, the parameterization of the prior and approximate posterior distributions are computed separately (Burda et al., 2015; Rezende et al., 2014). In contrast, Sønderby et al. (2016) propose a top-down dependency structure shared across the generative and inference models. From a probabilistic point of view, the approximate Gaussian likelihood, computed bottom-up by the inference model, is combined with the Gaussian prior, computed top-down from the generative model. We follow a similar procedure in computing the approximate posterior.

First, the parameters of the approximate likelihood are computed for each stochastic layer l:

[ˆµl t,q, ˆσl t,q] = f (l) q (zl+1 t , dl t) , (10) followed by the downward pass, recursively computing the prior and approximate posterior by precision-weighted addition:

σl t,q = 1 (ˆσl t,q) 2 + (σl t,p) 2 ,

µl t,q = σl t,q(ˆµl t,q(ˆσl t,q) 2 + µl t,p(σl t,p) 2) . (11)

Finally, the approximate posterior has the same decomposition as the prior (see Eq. (7)):

qφ(zt|x1:t) = qφ(z L t |d L t )

l=1 qφ(zl t|zl+1 t , dl t) , (12)

qφ(zl t|zl+1 t , dl t) = N(µl t,q, σl t,q) . (13)

Note that the inference and generative network share the parameters of dilated convolutions Conv(l).

3.3 LEARNING

The variational lower-bound on the log-likelihood at time-step t can be defined as follows: log p(xt) Eqφ(zt|xt)[log pθ(xt|zt)] DKL(qφ(zt|x1:t)||pθ(zt|x1:t 1))

= Eqφ(z1 t ...z L t |xt)[log pθ(xt|z1 t . . . z L t )] DKL(qφ(z1 t . . . z L t |x1:t)||pθ(z1 t . . . z L t |x1:t 1))

Lt(θ, φ; xt) = LRecon t + LKL t . (14)

Published as a conference paper at ICLR 2019

Using the decompositions from Eq. (7) and (12), the Kullback-Leibler divergence term becomes:

LKL t = DKL(qφ(z L t |d L t )||pθ(z L t |d L t 1))

l=1 Eqφ(zl+1 t | )[DKL(qφ(zl t|zl+1 t , dl t)||pθ(zl t|zl+1 t , dl t 1))] . (15)

The KL term is the same for the STCN and STCN-dense variants. The reconstruction term LRecon t , however, is different. In STCN we only use samples from the lowest layer of the hierarchy, whereas in STCN-dense we use all latent samples in the observation model:

LRecon t = Eqφ(z1 t ...z L t |xt)[log pθ(xt|z1 t )] , (16)

LRecon dense t = Eqφ(z1 t ...z L t |xt)[log pθ(xt|z1 t . . . z L t ] . (17)

In the dense variant, samples drawn from the latent variables zl t are carried over the dense connections. Similar to Maaløe et al. (2016), the expectation over zl t variables are computed by Monte Carlo sampling using the reparameterization trick (Kingma & Welling, 2013; Rezende et al., 2014).

Please note that the computation of LRecon dense t does not introduce any additional computational cost. In STCN, all latent variables have to be visited in terms of ancestral sampling in order to draw the latent sample z1 t for the observation xt. Similarly in STCN-dense, the same intermediate samples zl t are used in the prediction of xt.

One alternative option to use the latent samples could be to sum individual samples before feeding them into the observation model, i.e., sum([z1 t . . . z L t ]), (Maaløe et al., 2016). We empirically found that this does not work well in STCN-dense. Instead, we concatenate all samples [z1 t z L t ] analogously to Dense Net (Huang et al., 2017) and (Kaiser et al., 2018).

4 EXPERIMENTS

We evaluate the proposed variants STCN and STCN-dense both quantitatively and qualitatively on modeling of digital handwritten text and speech. We compare with vanilla TCNs, RNNs, VRNNs and state-of-the art models on the corresponding tasks.

In our experiments we use two variants of the Wavenet model: (1) the original model proposed in (Van Den Oord et al., 2016) and (2) a variant that we augment with skip connections analogously to STCN-dense. This additional baseline evaluates the benefit of learning multi-scale representations in the deterministic setting. Details of the experimental setup are provided in the Appendix. Our code is available at https://ait.ethz.ch/projects/2019/stcn/.

Handwritten text: The IAM-On DB and Deepwriting datasets consist of digital handwriting sequences where each time-step contains real-valued (x, y) pen coordinates and a binary pen-up event. The IAM-On DB data is split and pre-processed as done in (Chung et al., 2015). Aksan et al. (2018) extend this dataset with additional samples and better pre-processing.

Table 1 reveals that again both our variants outperform the vanilla variants of TCNs and RNNs on IAM-On DB. While the stochastic VRNN and SWave Net are competitive wrt to the STCN variant, both are outperformed by

(a) Ground truth

(c) Swave Net

(d) STCN-dense

Figure 3: (a) Handwriting samples from IAM-On DB dataset. Generated samples from (b) VRNN, (c) SWave Net and (d) our model STCN-dense. Each line corresponds to one sample.

Published as a conference paper at ICLR 2019

Table 1: Average log-likelihood per sequence on TIMIT, Blizzard, IAM-On DB and Deepwriting datasets. (Normal) and (GMM) stand for unimodal Gaussian or multi-modal Gaussian Mixture Model (GMM) as the observation model (Graves, 2013; Chung et al., 2015). Asterisks indicate that we used our re-implementation only for the Deepwriting dataset.

Models TIMIT Blizzard IAM-On DB Deepwriting Wavenet (GMM) 30188 8190 1381 612 Wavenet-dense (GMM) 30636 8212 1380 642 RNN (GMM) Chung et al. (2015) 26643 7413 1358 528

VRNN (Normal) Chung et al. (2015) 30235 9516 1354 495

VRNN (GMM) Chung et al. (2015) 29604 9392 1384 673

SRNN (Normal) Fraccaro et al. (2016) 60550 11991 n/a n/a Z-forcing (Normal) Goyal et al. (2017) 70469 15430 n/a n/a Var. Bi-LSTM (Normal) Shabanian et al. (2017) 73976 17319 n/a n/a SWave Net (Normal) Lai et al. (2018) 72463 15708 1301 n/a STCN (GMM) 69195 15800 1338 605 STCN-dense (GMM) 71386 16288 1796 797 STCN-dense-large (GMM) 77438 17670 n/a n/a

the STCN-dense version. The same relative ordering is maintained on the Deepwriting dataset, indicating that the proposed architecture is robust across datasets.

Fig. 3 compares generated handwriting samples. While all models produce consistent style, our model generates more natural looking samples. Note that the spacing between words is clearly visible and most of the letters are distinguishable.

Speech modeling: TIMIT and Blizzard are standard benchmark dataset in speech modeling. The models are trained and tested on 200 dimensional real-valued amplitudes. We apply the same pre-processing as Chung et al. (2015). For this task we introduce STCN-dense-large, with increased model capacity. Here we use 512 instead of 256 convolution filters. Note that the total number of model parameters is comparable to SWave Net and other SOA models.

On TIMIT, STCN-dense (Table 1) significantly outperforms the vanilla TCN and RNN, and stochastic models. On the Blizzard dataset, our model is marginally better than the Variational Bi-LSTM. Note that the inference models of SRNN (Fraccaro et al., 2016), Z-forcing (Goyal et al., 2017), and Variational Bi-LSTM (Shabanian et al., 2017) receive future information by using backward RNN cells. Similarly, SWave Net (Lai et al., 2018) applies causal convolutions in the backward direction. Hence, the latent variable can be expected to model future dynamics of the sequence. In contrast, our models have only access to information up to the current time-step. These results indicate that the STCN variants perform very well on the speech modeling task.

Latent Space Analysis: Zhao et al. (2017) observe that in hierarchical latent variable models the upper layers have a tendency to become inactive, indicated by a low KL loss (Sønderby et al., 2016; Dieng et al., 2018). Table 2 shows the KL loss per latent variable and the corresponding log-likelihood measured by ELBO in our models. Across the datasets it can be observed that our models make use of many of the latent variables which may explain the strong performance across tasks in terms of log-likelihoods. Note that STCN uses a standard hierarchical structure. However, individual latent variables have different information context due to the corresponding TCN block s receptive field. This observation suggests that the proposed combination of TCNs and stochastic variables is indeed effective. Furthermore, in STCN we see a similar utilization pattern of the z variables across tasks, whereas STCN-dense may have more flexibility in modeling the temporal dependencies within the data due to its dense connections to the output layer.

Table 2: KL-loss per latent variable computed over the entire test split. KL5 corresponds to the KL-loss of the top-most latent variable.

Dataset (Model) ELBO KL KL1 KL2 KL3 KL4 KL5 IAM-On DB (STCN-dense) 1796.3 1653.9 17.9 1287.4 305.3 41.0 2.4 IAM-On DB (STCN) 1339.2 964.2 846.0 105.2 12.9 0.1 0.0 TIMIT (STCN-dense) 71385.9 22297.5 16113.0 5641.6 529.0 8.3 5.7 TIMIT (STCN) 69194.9 23118.3 22275.5 487.2 355.5 0.0 0.0

Published as a conference paper at ICLR 2019

Replacing TCN with RNN: To better understand potential symergies between dilated CNNs and the proposed latent variable hierarchy, we perform an ablation study, isolating the effect of TCNs and the latent space. To this end the deterministic TCN blocks are replaced with LSTM cells by keeping the latent structure intact. We dub this condition Ladder RNN. We use the TIMIT and IAM-On DB datasets for evaluation. Table 3 summarizes performance measured by the ELBO.

The most direct translation of the the STCN architecture into an RNN counterpart has 25 stacked LSTM cells with 256 units each. Similar to STCN, we use 5 stochastic layers (see Appendix 7.1). Note that stacking this many LSTM cells is unusual and resulted in instabilities during training. Hence, the performance is similar to vanilla RNNs. The second Ladder RNN configuration uses 5 stacked LSTM cells with 512 units and a one-to-one mapping with the stochastic layers. On the TIMIT dataset, all Ladder RNN configurations show a significant improvement. We also observe a pattern of improvement with densely connected latent variables.

This experiments shows that the proposed modular latent variable design does allow for the usage of different building blocks. Even when attached to LSTM cells, it boosts the log-likelihood performance (see 5x512Ladder RNN), in particular when used with dense connections. However, the empirical results suggest that the densely connected latent hierarchy interacts particularly well with dilated CNNs. We suggest this is due to the hierarchical nature on both sides of the architecture. On both datasets STCN models achieved the best performance and significantly improve with dense connections. This supports our contribution of a latent variable hierarchy, which models different aspects of information from the input time-series.

Table 3: ELBO of Ladder RNN and STCN models using the same latent space configuration. The prefix of a model entries denote the number of RNN or TCN layers and unit size per layer. Models have similar number of trainable parameters.

Models TIMIT IAM-On DB 25x256-Ladder RNN (Normal) 28207 1305 25x256-Ladder RNN-dense (Normal) 27413 1278 25x256-Ladder RNN (GMM) 24839 1381 25x256-Ladder RNN-dense (GMM) 26240 1377 5x512-Ladder RNN (Normal) 49770 1299 5x512-Ladder RNN-dense (Normal) 48612 1374 5x512-Ladder RNN (GMM) 47179 1359 5x512-Ladder RNN-dense (GMM) 50113 1581 25x256-STCN (Normal) 64913 1327 25x256-STCN-dense (Normal) 70294 1729 25x256-STCN (GMM) 69195 1339 25x256-STCN-dense (GMM) 71386 1796

5 RELATED WORK

Rezende et al. (2014) propose Deep Latent Gaussian Models (DLGM) and Sønderby et al. (2016) propose the Ladder Variational Autoencoder (LVAE). In both models the latent variables are hierarchically defined and conditioned on the preceding stochastic layer. LVAEs improve upon DLGMs via implementation of a top-down hierarchy both in the generative and inference model. The approximate posterior is computed via a precisionweighted update of the approximate likelihood (i.e., the inference model) and prior (i.e., the generative model). Similarly, the Pixel VAE (Gulrajani et al., 2016) incorporates a hierarchical latent space decomposition and uses an autoregressive decoder. Zhao et al. (2017) show under mild conditions that straightforward stacking of latent variables (as is done e.g. in LVAE and Pixel VAE) can be ineffective, because the latent variables that are not directly conditioned on the observation variable become inactive.

Due to the nature of the sequential problem domain, our approach differs in the crucial aspects that STCNs use dynamic, i.e., conditional, priors (Chung et al., 2015) at every level. Moreover, the hierarchy is not only implicitly defined by the network architecture but also explicitly defined by the information content, i.e., receptive field size. Dieng et al. (2018) both theoretically and empirically show that using skip connections from the latent variable to every layer of the decoder increases mutual information between the latent and observation variables. Similar to Dieng et al. (2018) in STCN-dense, we introduce skip connections from all latent variables to the output. In STCN the model is expected to encode and propagate the information through its hierarchy.

Yang et al. (2017) suggest using autoregressive TCN decoders to remedy the posterior collapse problem observed in language modeling with LSTM decoders (Bowman et al., 2015). van den Oord et al. (2017) and Dieleman et al. (2018) use TCN decoders conditioned on discrete latent variables to model audio signals.

Published as a conference paper at ICLR 2019

Stochastic RNN architectures mostly vary in the way they employ the latent variable and parametrize the approximate posterior for variational inference. Chung et al. (2015) and Bayer & Osendorfer (2014) use the latent random variable to capture high-level information causing the variability observed in sequential data. Particularly Chung et al. (2015) shows that using a conditional prior rather than a standard Gaussian distribution is very effective in sequence modeling. In (Fraccaro et al., 2016; Goyal et al., 2017; Shabanian et al., 2017), the inference model, i.e., the approximate posterior, receives both the past and future summaries of the sequence from the hidden states of forward and backward RNN cells. The KL-divergence term in the objective enforces the model to learn predictive latent variables in order to capture the future states of the sequence.

Lai et al. (2018) s SWave Net is most closely related to ours. SWave Net also introduces latent variables into TCNs. However, in SWave Net the deterministic and stochastic units are coupled which may prevent stacking of larger numbers of TCN blocks. Since the number of stacked dilated convolutions determines the receptive field size, this directly correlates with the model capacity. For example, the performance of SWave Net on the IAM-On DB dataset degrades after stacking more than 3 stochastic layers (Lai et al., 2018), limiting the model to a small receptive field. In contrast, we aim to preserve the flexibility of stacking dilated convolutions in the base TCN. In STCNs, the deterministic TCN units do not have any dependency on the stochastic variables (see Figure 1) and the ratio of stochastic to deterministic units can be adjusted, depending on the task.

6 CONCLUSION

In this paper we proposed STCNs, a novel auto-regressive model, combining the computational benefits of convolutional architectures and expressiveness of hierarchical stochastic latent spaces. We have shown the effectivness of the approach across several sequence modelling tasks and datasets. The proposed models are trained via optimization of the ELBO objective. Tighter lower bounds such as IWAE (Burda et al., 2015) or FIVO (Maddison et al., 2017) may further improve modeling performance. We leave this for future work.

ACKNOWLEDGEMENTS

This work was supported in parts by the ERC grant OPTINT (St G-2016-717054). We gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this research.

Published as a conference paper at ICLR 2019

Martin Abadi, Paul Barham, Jianmin Chen, Zhifeng Chen, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Geoffrey Irving, Michael Isard, Manjunath Kudlur, Josh Levenberg, Rajat Monga, Sherry Moore, Derek G. Murray, Benoit Steiner, Paul Tucker, Vijay Vasudevan, Pete Warden, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. Tensorflow: A system for large-scale machine learning. In 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16), pp. 265 283, 2016. URL https://www.usenix.org/system/files/conference/osdi16/osdi16-abadi.pdf.

Emre Aksan, Fabrizio Pece, and Otmar Hilliges. Deep Writing: Making Digital Ink Editable via Deep Generative Modeling. In SIGCHI Conference on Human Factors in Computing Systems, CHI 18, New York, NY, USA, 2018. ACM.

Shaojie Bai, J Zico Kolter, and Vladlen Koltun. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. ar Xiv preprint ar Xiv:1803.01271, 2018.

Justin Bayer and Christian Osendorfer. Learning stochastic recurrent networks. ar Xiv preprint ar Xiv:1411.7610, 2014.

Samuel R Bowman, Luke Vilnis, Oriol Vinyals, Andrew M Dai, Rafal Jozefowicz, and Samy Bengio. Generating sentences from a continuous space. ar Xiv preprint ar Xiv:1511.06349, 2015.

Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. ar Xiv preprint ar Xiv:1509.00519, 2015.

Kyunghyun Cho, Bart Van Merri enboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. Learning phrase representations using rnn encoder-decoder for statistical machine translation. ar Xiv preprint ar Xiv:1406.1078, 2014.

Junyoung Chung, Kyle Kastner, Laurent Dinh, Kratarth Goel, Aaron C Courville, and Yoshua Bengio. A recurrent latent variable model for sequential data. In Advances in neural information processing systems, pp. 2980 2988, 2015.

Yann N Dauphin, Angela Fan, Michael Auli, and David Grangier. Language modeling with gated convolutional networks. ar Xiv preprint ar Xiv:1612.08083, 2016.

Sander Dieleman, A aron van den Oord, and Karen Simonyan. The challenge of realistic music generation: modelling raw audio at scale. ar Xiv preprint ar Xiv:1806.10474, 2018.

Adji B Dieng, Yoon Kim, Alexander M Rush, and David M Blei. Avoiding latent variable collapse with generative skip models. ar Xiv preprint ar Xiv:1807.04863, 2018.

Otto Fabius and Joost R van Amersfoort. Variational recurrent auto-encoders. ar Xiv preprint ar Xiv:1412.6581, 2014.

Marco Fraccaro, Søren Kaae Sønderby, Ulrich Paquet, and Ole Winther. Sequential neural models with stochastic layers. In Advances in neural information processing systems, pp. 2199 2207, 2016.

Jonas Gehring, Michael Auli, David Grangier, Denis Yarats, and Yann N Dauphin. Convolutional sequence to sequence learning. ar Xiv preprint ar Xiv:1705.03122, 2017.

Anirudh Goyal ALIAS PARTH Goyal, Alessandro Sordoni, Marc-Alexandre Cˆot e, Nan Ke, and Yoshua Bengio. Z-forcing: Training stochastic recurrent networks. In Advances in Neural Information Processing Systems, pp. 6713 6723, 2017.

Alex Graves. Generating sequences with recurrent neural networks. ar Xiv preprint ar Xiv:1308.0850, 2013.

Ishaan Gulrajani, Kundan Kumar, Faruk Ahmed, Adrien Ali Taiga, Francesco Visin, David Vazquez, and Aaron Courville. Pixelvae: A latent variable model for natural images. ar Xiv preprint ar Xiv:1611.05013, 2016.

Sepp Hochreiter and J urgen Schmidhuber. Long short-term memory. Neural computation, 9(8):1735 1780, 1997.

Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. Densely connected convolutional networks. In CVPR, volume 1, pp. 3, 2017.

Łukasz Kaiser, Aurko Roy, Ashish Vaswani, Niki Pamar, Samy Bengio, Jakob Uszkoreit, and Noam Shazeer. Fast decoding in sequence models using discrete latent variables. ar Xiv preprint ar Xiv:1803.03382, 2018.

Diederik P Kingma and Max Welling. Auto-encoding variational bayes. ar Xiv preprint ar Xiv:1312.6114, 2013.

Published as a conference paper at ICLR 2019

Guokun Lai, Bohan Li, Guoqing Zheng, and Yiming Yang. Stochastic wavenet: A generative latent variable model for sequential data, 2018.

Lars Maaløe, Casper Kaae Sønderby, Søren Kaae Sønderby, and Ole Winther. Auxiliary deep generative models. ar Xiv preprint ar Xiv:1602.05473, 2016.

Chris J Maddison, John Lawson, George Tucker, Nicolas Heess, Mohammad Norouzi, Andriy Mnih, Arnaud Doucet, and Yee Teh. Filtering variational objectives. In Advances in Neural Information Processing Systems, pp. 6573 6583, 2017.

Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and approximate inference in deep generative models. ar Xiv preprint ar Xiv:1401.4082, 2014.

Samira Shabanian, Devansh Arpit, Adam Trischler, and Yoshua Bengio. Variational bi-lstms. ar Xiv preprint ar Xiv:1711.05717, 2017.

Casper Kaae Sønderby, Tapani Raiko, Lars Maaløe, Søren Kaae Sønderby, and Ole Winther. Ladder variational autoencoders. In Advances in neural information processing systems, pp. 3738 3746, 2016.

A aron Van Den Oord, Sander Dieleman, Heiga Zen, Karen Simonyan, Oriol Vinyals, Alex Graves, Nal Kalchbrenner, Andrew W Senior, and Koray Kavukcuoglu. Wavenet: A generative model for raw audio. In SSW, pp. 125, 2016.

A aron van den Oord, Nal Kalchbrenner, Lasse Espeholt, Oriol Vinyals, Alex Graves, et al. Conditional image generation with pixelcnn decoders. In Advances in Neural Information Processing Systems, pp. 4790 4798, 2016.

Aaron van den Oord, Oriol Vinyals, et al. Neural discrete representation learning. In Advances in Neural Information Processing Systems, pp. 6306 6315, 2017.

Zichao Yang, Zhiting Hu, Ruslan Salakhutdinov, and Taylor Berg-Kirkpatrick. Improved variational autoencoders for text modeling using dilated convolutions. ar Xiv preprint ar Xiv:1702.08139, 2017.

Fisher Yu and Vladlen Koltun. Multi-scale context aggregation by dilated convolutions. ar Xiv preprint ar Xiv:1511.07122, 2015.

Shengjia Zhao, Jiaming Song, and Stefano Ermon. Learning hierarchical features from generative models. ar Xiv preprint ar Xiv:1702.08396, 2017.

Published as a conference paper at ICLR 2019

7.1 NETWORK DETAILS

Figure 4: Generative model of STCN-dense architecture. Building blocks are highlighted. Note that the dependence of dl t, l = 1 L on past inputs is not visualized for clarity.

The network architecture of the proposed model is illustrated in Fig. 4. We make only a small modification to the vanilla Wavenet architecture. Instead of using skip connections from Wavenet blocks, we only use the latent sample zt in order to make a prediction of xt. In STCN-dense configuration, zt is the concatenation of all latent variables in the hierarchy, i.e., zt = [z1 t z L t ], whereas in STCN only z1 t is fed to the output layer.

Each stochastic latent variable zl t (except the top-most z L t ) is conditioned on a deterministic TCN representation dl t and the preceding random variable zl+1 t . The latent variables are calculated by using the latent layers f (l) p or f (l) q which are neural networks.

We do not define a latent variable per TCN layer. Instead, the stochastic layers are uniformly distributed where each random variable is conditioned on a number of stacked TCN layers dl t. We stack K Wavenet blocks (see figure 4 left) with exponentially increasing dilation size.

Observation Model: We use Normal or GMM distributions with 20 components to model real-valued data. All Gaussian distributions have diagonal covariance matrix.

Output layer f (o): For the IAM-On DB and Deepwriting datasets we use 1D convolutions with Re LU nonlinearity. We stack 5 of these layers with 256 filters and filter size 1.

For TIMIT and Blizzard datasets Wavenet blocks in the output layer perform significantly better. We stack 5 Wavenet blocks with dilation size 1. For each convolution operation in the block we use 256 filters. The filter size of the dilated convolution is set to 2. The STCN-dense-large model is constructed by using 512 filters instead of 256.

TCN blocks dl t: The number of Wavenet blocks is usually determined by the desired receptive field size.

For the handwriting datasets K = 6 and L = 5. In total we have 30 Wavenet blocks where each convolution operation has 256 filters with size 2.

For speech datasets K = 5 and L = 5. In total we have 25 Wavenet blocks where each convolution operation has 256 filters with size 2. The large model configuration uses 512 filters.

Latent layers f (l) p and f (l) q : The number of stochastic layers per task is given by L. We used [32, 16, 8, 5, 2] dimensional latent variables for the handwriting tasks. It is [256, 128, 64, 32, 16] for speech datasets. Note that the first entry of the list corresponds to z1.

The mean and sigma parameters of the Normal distributions modeling the latent variables are calculated by the f (l) p and f (l) q networks. We stack 2 1D convolutions with Re LU nonlinearity and filter size 1. The number of filters are the same as the number of Wavenet block filters for the corresponding task.

Finally, we clamped the latent sigma predictions between 0.001 and 5.

Published as a conference paper at ICLR 2019

7.2 TRAINING DETAILS

In all STCN experiments we applied KL annealing. In all tasks, the weight of the KL term is initialized with 0 and increased by 1 e 4 at every step until it reaches 1.

The batch size was 20 for all datasets except for Blizzard where it was 128.

We use the ADAM optimizer with its default parameters and exponentially decay the learning rate. For the handwriting datasets the learning rate was initialized with 5 e 4 and followed a decay rate of 0.94 over 1000 decay steps. On the speech datasets it was initialized with 1 e 3 and decayed with a rate of 0.98. We applied early stopping by measuring the ELBO performance on the validation splits.

We implement STCN models in Tensorflow (Abadi et al., 2016). Our code and models achieving the SOA results are available at https://ait.ethz.ch/projects/2019/stcn/.

7.3 DETAILED RESULTS

Here we provide the extended results table with Normal observation model entries for available models.

Table 4: Average log-likelihood per sequence on TIMIT, Blizzard, IAM-On DB and Deepwriting datasets. (Normal) and (GMM) stand for unimodal Gaussian or multi-modal Gaussian Mixture Model (GMM) as the observation model (Graves, 2013; Chung et al., 2015). Asterisks indicate that we used our re-implementation only for the Deepwriting dataset.

Models TIMIT Blizzard IAM-On DB Deepwriting Wavenet (Normal) -7443 3784 1053 337 Wavenet (GMM) 30188 8190 1381 612 Wavenet-dense (Normal) -8579 3712 1030 323 Wavenet-dense (GMM) 30636 8212 1380 642 RNN (Normal) Chung et al. (2015) -1900 3539 1016 363

RNN (GMM) Chung et al. (2015) 26643 7413 1358 528

VRNN (Normal)Chung et al. (2015) 30235 9516 1354 495

VRNN (GMM) Chung et al. (2015) 29604 9392 1384 673

SRNN (Normal) Fraccaro et al. (2016) 60550 11991 n/a n/a Z-forcing (Normal)Goyal et al. (2017) 70469 15430 n/a n/a Var. Bi-LSTM (Normal)Shabanian et al. (2017) 73976 17319 n/a n/a SWave Net (Normal)Lai et al. (2018) 72463 15708 1301 n/a STCN(Normal) 64913 13273 1327 575 STCN(GMM) 69195 15800 1338 605 STCN-dense(Normal) 70294 15950 1729 740 STCN-dense(GMM) 71386 16288 1796 797 STCN-dense-large (GMM) 77438 17670 n/a n/a