147

Recent contrastive representation learning methods rely on estimating mutual information (MI) between multiple views of an underlying context. E.g., we can derive multiple views of a given image by applying data augmentation, or we can split a sequence into views comprising the past and future of some step in the sequence. Contrastive lower bounds on MI are easy to optimize, but have a strong underestimation bias when estimating large amounts of MI. We propose decomposing the full MI estimation problem into a sum of smaller estimation problems by splitting one of the views into progressively more informed subviews and by applying the chain rule on MI between the decomposed views. This expression contains a sum of unconditional and conditional MI terms, each measuring modest chunks of the total MI, which facilitates approximation via contrastive bounds. To maximize the sum, we formulate a contrastive lower bound on the conditional MI which can be approximated efficiently. We refer to our general approach as Decomposed Estimation of Mutual Information (DEMI). We show that DEMI can capture a larger amount of MI than standard non-decomposed contrastive bounds in a synthetic setting, and learns better representations in a vision domain and for dialogue generation.

This paper introduces the Eighth Dialog System Technology Challenge. In line with recent challenges, the eighth edition focuses on applying end-to-end dialog technologies in a pragmatic way for multi-domain task-completion, noetic response selection, audio visual scene-aware dialog, and schema-guided dialog state tracking tasks. This paper describes the task definition, provided datasets, baselines and evaluation set-up for each track. We also summarize the results of the submitted systems to highlight the overall trends of the state-of-the-art technologies for the tasks.

Domain adaptation has recently become a key problem in dialogue systems research. Deep learning, while being the preferred technique for modeling such systems, works best given massive training data. However, in real-world scenarios, such resources are rarely available for new domains, and the ability to train with a few dialogue examples can be considered essential. Pre-training on large data sources and adapting to the target data has become the standard method for few-shot problems within the deep learning framework. In this paper, we presentgrtr, a hybrid generative-retrieval model based on the large-scale general-purpose language model GPT[2] fine-tuned to the multi-domainmetalwoz dataset. In addition to robust and diverse response generation provided by the GPT[2], our model is able to estimate generation confidence, and is equipped with retrieval logic as a fallback for the cases when the estimate is low.grtr is the winning entry at the fast domain adaptation task of DSTC-8 in human evaluation ($>$4% improvement over the 2nd place system). It also attains superior performance to a series of baselines on automated metrics onmetalwoz andmultiwoz, a multi-domain dataset of goal-oriented dialogues. In this paper, we also conduct a study ofgrtr's performance in the setup of limited adaptation data, evaluating the model's overall response prediction performance onmetalwoz and goal-oriented performance onmultiwoz.

Goal-oriented dialogue systems are now widely adopted in industry, where practical aspects of using them becomes of key importance. As such, it is expected from such systems to fit into a rapid prototyping cycle for new products and domains. For data-driven dialogue systems (especially those based on deep learning) that amounts to maintaining production-level performance having been provided with a few `seed' dialogue examples, normally referred to as data efficiency.With extremely data-dependent deep learning methods, the most promising way to achieve practical data efficiency is transfer learning-i.e., leveraging a greater, highly represented data source for training a base model, then fine-tuning it to available in-domain data.In this paper, we present a hybrid generative-retrieval model that can be trained using transfer learning. By using GPT-2 as the base model and fine-tuning it to the multidomain MetaLWOz dataset, we obtain a robust dialogue model able to perform both response generation and ranking1. Combining both, it outperforms several competitive generative-only and retrieval-only baselines, measured by language modeling quality on MetaLWOz as well as in goal- oriented metrics (Intent/Slot Fl-scores) on the MultiWoz corpus.

We introduce a new benchmark for coreference resolution and NLI, KnowRef, that targets common-sense understanding and world knowledge. Previous coreference resolution tasks can largely be solved by exploiting the number and gender of the antecedents, or have been handcrafted and do not reflect the diversity of naturally occurring text. We present a corpus of over 8,000 annotated text passages with ambiguous pronominal anaphora. These instances are both challenging and realistic. We show that various coreference systems, whether rule-based, feature-rich, or neural, perform significantly worse on the task than humans, who display high inter-annotator agreement. To explain this performance gap, we show empirically that state-of-the art models often fail to capture context, instead relying on the gender or number of candidate antecedents to make a decision. We then use problem-specific insights to propose a data-augmentation trick called antecedent switching to alleviate this tendency in models. Finally, we show that antecedent switching yields promising results on other tasks as well: we use it to achieve state-of-the-art results on the GAP coreference task.

Conditional text-to-image generation is an active area of research, with many possible applications. Existing research has primarily focused on generating a single image from available conditioning information in one step. One practical extension beyond one-step generation is a system that generates an image iteratively, conditioned on ongoing linguistic input or feedback. This is significantly more challenging than one-step generation tasks, as such a system must understand the contents of its generated images with respect to the feedback history, the current feedback, as well as the interactions among concepts present in the feedback history. In this work, we present a recurrent image generation model which takes into account both the generated output up to the current step as well as all past instructions for generation. We show that our model is able to generate the background, add new objects, and apply simple transformations to existing objects. We believe our approach is an important step toward interactive generation. Code and data is available at: https://www.microsoft.com/en-us/research/project/generative-neural-visual-artist-geneva/.

There has been growing interest in using neural networks and deep learning techniques to create dialogue systems. Conversational recommendation is an interesting setting for the scientific exploration of dialogue with natural language as the associated discourse involves goal-driven dialogue that often transforms naturally into more free-form chat. This paper provides two contributions. First, until now there has been no publicly available large-scale data set consisting of real-world dialogues centered around recommendations. To address this issue and to facilitate our exploration here, we have collected ReDial, a data set consisting of over 10,000 conversations centered around the theme of providing movie recommendations. We make this data available to the community for further research. Second, we use this dataset to explore multiple facets of conversational recommendations. In particular we explore new neural architectures, mechanisms and methods suitable for composing conversational recommendation systems. Our dataset allows us to systematically probe model sub-components addressing different parts of the overall problem domain ranging from: sentiment analysis and cold-start recommendation generation to detailed aspects of how natural language is used in this setting in the real world. We combine such sub-components into a full-blown dialogue system and examine its behavior.

This paper proposes a new dataset, Frames, composed of 1369 human-human dialogues with an average of 15 turns per dialogue.This corpus contains goal-oriented dialogues between users who are given some constraints to book a trip and assistants who search a database to find appropriate trips.The users exhibit complex decision-making behaviour which involve comparing trips, exploring different options, and selecting among the trips that were discussed during the dialogue.To drive research on dialogue systems towards handling such behaviour, we have annotated and released the dataset and we propose in this paper a task called frame tracking.This task consists of keeping track of different semantic frames throughout each dialogue.We propose a rule-based baseline and analyse the frame tracking task through this baseline.

Reliable object discovery in realistic indoor scenes is a necessity for many computer vision and service robot applications. In these scenes, semantic segmentation methods have made huge advances in recent years. Such methods can provide useful prior information for object discovery by removing false positives and by delineating object boundaries. We propose a novel method that combines bottom-up object discovery and semantic priors for producing generic object candidates in RGB-D images. We use a deep learning method for semantic segmentation to classify colour and depth superpixels into meaningful categories. Separately for each category, we use saliency to estimate the location and scale of objects, and superpixels to find their precise boundaries. Finally, object candidates of all categories are combined and ranked. We evaluate our approach on the NYU Depth V2 dataset and show that we outperform other state-of-the-art object discovery methods in terms of recall.

In this paper, we propose to use deep policy networks which are trained with an advantage actor-critic method for statistically optimised dialogue systems. First, we show that, on summary state and action spaces, deep Reinforcement Learning (RL) outperforms Gaussian Processes methods. Summary state and action spaces lead to good performance but require pre-engineering effort, RL knowledge, and domain expertise. In order to remove the need to define such summary spaces, we show that deep RL can also be trained efficiently on the original state and action spaces. Dialogue systems based on partially observable Markov decision processes are known to require many dialogues to train, which makes them unappealing for practical deployment. We show that a deep RL method based on an actor-critic architecture can exploit a small amount of data very efficiently. Indeed, with only a few hundred dialogues collected with a handcrafted policy, the actor-critic deep learner is considerably bootstrapped from a combination of supervised and batch RL. In addition, convergence to an optimal policy is significantly sped up compared to other deep RL methods initialized on the data with batch RL. All experiments are performed on a restaurant domain derived from the Dialogue State Tracking Challenge 2 (DSTC2) dataset.

Object recognition and pose estimation from RGB-D images are important tasks for manipulation robots which can be learned from examples. Creating and annotating datasets for learning is expensive, however. We address this problem with transfer learning from deep convolutional neural networks (CNN) that are pre-trained for image categorization and provide a rich, semantically meaningful feature set. We incorporate depth information, which the CNN was not trained with, by rendering objects from a canonical perspective and colorizing the depth channel according to distance from the object center. We evaluate our approach on the Washington RGB-D Objects dataset, where we find that the generated feature set naturally separates classes and instances well and retains pose manifolds. We outperform state-of-the-art on a number of subtasks and show that our approach can yield superior results when only little training data is available.

Object-class segmentation is a computer vision task which requires labeling each pixel of an image with the class of the object it belongs to. Deep convolutional neural networks (DNN) are able to learn and exploit local spatial correlations required for this task. They are, however, restricted by their small, fixed-sized filters, which limits their ability to learn long-range dependencies. Recurrent Neural Networks (RNN), on the other hand, do not suffer from this restriction. Their iterative interpretation allows them to model long-range dependencies by propagating activity. This property might be especially useful when labeling video sequences, where both spatial and temporal long-range dependencies occur. In this work, we propose novel RNN architectures for object-class segmentation. We investigate three ways to consider past and future context in the prediction process by comparing networks that process the frames one by one with networks that have access to the whole sequence. We evaluate our models on the challenging NYU Depth v2 dataset for object-class segmentation and obtain competitive results.

We developed a robust real time hand gesture based interaction system to effectively communicate with a mobile robot which can operate in an outdoor environment. The system enables the user to operate a mobile robot using hand gesture based commands. In particular the system offers direct on site interaction providing better perception of environment to the user. To overcome the illumination challenges in outdoors, the system operates on depth images. Processed depth images are given as input to a convolutional neural network which is trained to detect static hand gestures.

Unsupervised learning algorithms find ways to model latent structure present in the data. These latent structures can then serve as a basis for supervised classification methods. A common choice for unsupervised feature discovery is the Restricted Boltzmann Machine (RBM). Since the RBM is a general purpose learning machine, it is not particularly tailored for image data. Representations found by RBMs are consequently not image-like. Since it is essential to exploit the known topological structure for image analysis, it is desirable not to discard the topology property when learning new representations. Then, the same learning methods can be applied to the latent representation in a hierarchical manner. In this work, we propose a modification to the learning rule of locally connected RBMs, which ensures that topological image structure is preserved in the latent representation. To this end, we use a Gaussian kernel to transfer topological properties of the image space to the feature space. The learned model is then used as an initialization for a neural network trained to classify the images. We evaluate our approach on the MNIST and Caltech 101 datasets and demonstrate that we are able to learn topological feature maps.

Recently, multithreading became a standard technique to improve the processor utilization and system performance. Hardware support is provided for coarse-grained as well as simultaneous multithreading. In particular, embedded devices combine processor cores and varying sets of coprocessors to fulfill the requirements of their dedicated application field. In this paper, a simultaneous multithreaded processor is investigated that applies dynamic priorities for each thread on the instruction level. By means of a synchronization coprocessor, priorities of threads are dynamically adapted when other threads have to wait for a given thread. Based on simulations of a network-processing workload, two strategies of dynamic priority adaptation are evaluated and compared with static prioritization. As a result, performance gain can be shown.