Search

Inspired by the success of the Transformer architecture in natural language processing and computer vision, we investigate the use of Transformers in Reinforcement Learning (RL), specifically in modeling the environment's dynamics using Transformer Dynamics Models (TDMs). We evaluate the capabilities of TDMs for continuous control in real-time planning scenarios with Model Predictive Control (MPC). While Transformers excel in long-horizon prediction, their tokenization mechanism and autoregressive nature lead to costly planning over long horizons, especially as the environment's dimensionality increases. To alleviate this issue, we use a TDM for short-term planning, and learn an autoregressive discrete Q-function using a separate Q-Transformer (QT) model to estimate a long-term return beyond the short-horizon planning. Our proposed method, QT-TDM, integrates the robust predictive capabilities of Transformers as dynamics models with the efficacy of a model-free Q-Transformer to mitigate the computational burden associated with real-time planning. Experiments in diverse state-based continuous control tasks show that QT-TDM is superior in performance and sample efficiency compared to existing Transformer-based RL models while achieving fast and computationally efficient inference.

Language-conditioned robotic skills make it possible to apply the high-level reasoning of Large Language Models (LLMs) to low-level robotic control. A remaining challenge is to acquire a diverse set of fundamental skills. Existing approaches either manually decompose a complex task into atomic robotic actions in a top-down fashion, or bootstrap as many combinations as possible in a bottom-up fashion to cover a wider range of task possibilities. These decompositions or combinations, however, require an initial skill library. For example, a ``grasping'' capability can never emerge from a skill library containing only diverse ``pushing'' skills. Existing skill discovery techniques with reinforcement learning acquire skills by an exhaustive exploration but often yield non-meaningful behaviors. In this study, we introduce a novel framework for skill discovery that is entirely driven by LLMs. The framework begins with an LLM generating task proposals based on the provided scene description and the robot's configurations, aiming to incrementally acquire new skills upon task completion. For each proposed task, a series of reinforcement learning processes are initiated, utilizing reward and success determination functions sampled by the LLM to develop the corresponding policy. The reliability and trustworthiness of learned behaviors are further ensured by an independent vision-language model. We show that starting with zero skill, the skill library emerges and expands to more and more meaningful and reliable skills, enabling the robot to efficiently further propose and complete advanced tasks. Project page: \url{https://agentic-skill-discovery.github.io}., Comment: Webpage see https://agentic-skill-discovery.github.io/

Although there has been rapid progress in endowing robots with the ability to solve complex manipulation tasks, generating control policies for bimanual robots to solve tasks involving two hands is still challenging because of the difficulties in effective temporal and spatial coordination. With emergent abilities in terms of step-by-step reasoning and in-context learning, Large Language Models (LLMs) have taken control of a variety of robotic tasks. However, the nature of language communication via a single sequence of discrete symbols makes LLM-based coordination in continuous space a particular challenge for bimanual tasks. To tackle this challenge for the first time by an LLM, we present LAnguage-model-based Bimanual ORchestration (LABOR), an agent utilizing an LLM to analyze task configurations and devise coordination control policies for addressing long-horizon bimanual tasks. In the simulated environment, the LABOR agent is evaluated through several everyday tasks on the NICOL humanoid robot. Reported success rates indicate that overall coordination efficiency is close to optimal performance, while the analysis of failure causes, classified into spatial and temporal coordination and skill selection, shows that these vary over tasks. The project website can be found at http://labor-agent.github.io, Comment: The project website can be found at http://labor-agent.github.io

Reinforcement Learning (RL) plays an important role in the robotic manipulation domain since it allows self-learning from trial-and-error interactions with the environment. Still, sample efficiency and reward specification seriously limit its potential. One possible solution involves learning from expert guidance. However, obtaining a human expert is impractical due to the high cost of supervising an RL agent, and developing an automatic supervisor is a challenging endeavor. Large Language Models (LLMs) demonstrate remarkable abilities to provide human-like feedback on user inputs in natural language. Nevertheless, they are not designed to directly control low-level robotic motions, as their pretraining is based on vast internet data rather than specific robotics data. In this paper, we introduce the Lafite-RL (Language agent feedback interactive Reinforcement Learning) framework, which enables RL agents to learn robotic tasks efficiently by taking advantage of LLMs' timely feedback. Our experiments conducted on RLBench tasks illustrate that, with simple prompt design in natural language, the Lafite-RL agent exhibits improved learning capabilities when guided by an LLM. It outperforms the baseline in terms of both learning efficiency and success rate, underscoring the efficacy of the rewards provided by an LLM., Comment: CoRL 2023 Workshop (oral)

Recent advancements in large language models have showcased their remarkable generalizability across various domains. However, their reasoning abilities still have significant room for improvement, especially when confronted with scenarios requiring multi-step reasoning. Although large language models possess extensive knowledge, their reasoning often fails to effectively utilize this knowledge to establish a coherent thinking paradigm. These models sometimes show hallucinations as their reasoning procedures are unconstrained by logical principles. Aiming at improving the zero-shot chain-of-thought reasoning ability of large language models, we propose LoT (Logical Thoughts), a self-improvement prompting framework that leverages principles rooted in symbolic logic, particularly Reductio ad Absurdum, to systematically verify and rectify the reasoning processes step by step. Experimental evaluations conducted on language tasks in diverse domains, including arithmetic, commonsense, symbolic, causal inference, and social problems, demonstrate the efficacy of enhanced reasoning by logic. The implementation code for LoT can be accessed at: https://github.com/xf-zhao/LoT., Comment: Accepted in COLING 2024. Code see https://github.com/xf-zhao/LoT

Programming robot behavior in a complex world faces challenges on multiple levels, from dextrous low-level skills to high-level planning and reasoning. Recent pre-trained Large Language Models (LLMs) have shown remarkable reasoning ability in few-shot robotic planning. However, it remains challenging to ground LLMs in multimodal sensory input and continuous action output, while enabling a robot to interact with its environment and acquire novel information as its policies unfold. We develop a robot interaction scenario with a partially observable state, which necessitates a robot to decide on a range of epistemic actions in order to sample sensory information among multiple modalities, before being able to execute the task correctly. Matcha (Multimodal environment chatting) agent, an interactive perception framework, is therefore proposed with an LLM as its backbone, whose ability is exploited to instruct epistemic actions and to reason over the resulting multimodal sensations (vision, sound, haptics, proprioception), as well as to plan an entire task execution based on the interactively acquired information. Our study demonstrates that LLMs can provide high-level planning and reasoning skills and control interactive robot behavior in a multimodal environment, while multimodal modules with the context of the environmental state help ground the LLMs and extend their processing ability. The project website can be found at https://matcha-agent.github.io., Comment: IROS2023, Detroit. See the project website at https://matcha-agent.github.io

Model-based reinforcement learning (MBRL) with real-time planning has shown great potential in locomotion and manipulation control tasks. However, the existing planning methods, such as the Cross-Entropy Method (CEM), do not scale well to complex high-dimensional environments. One of the key reasons for underperformance is the lack of exploration, as these planning methods only aim to maximize the cumulative extrinsic reward over the planning horizon. Furthermore, planning inside the compact latent space in the absence of observations makes it challenging to use curiosity-based intrinsic motivation. We propose Curiosity CEM (CCEM), an improved version of the CEM algorithm for encouraging exploration via curiosity. Our proposed method maximizes the sum of state-action Q values over the planning horizon, in which these Q values estimate the future extrinsic and intrinsic reward, hence encouraging reaching novel observations. In addition, our model uses contrastive representation learning to efficiently learn latent representations. Experiments on image-based continuous control tasks from the DeepMind Control suite show that CCEM is by a large margin more sample-efficient than previous MBRL algorithms and compares favorably with the best model-free RL methods., Comment: 7 pages, 4 figures

Due to the dynamic nature of human language, automatic speech recognition (ASR) systems need to continuously acquire new vocabulary. Out-Of-Vocabulary (OOV) words, such as trending words and new named entities, pose problems to modern ASR systems that require long training times to adapt their large numbers of parameters. Different from most previous research focusing on language model post-processing, we tackle this problem on an earlier processing level and eliminate the bias in acoustic modeling to recognize OOV words acoustically. We propose to generate OOV words using text-to-speech systems and to rescale losses to encourage neural networks to pay more attention to OOV words. Specifically, we enlarge the classification loss used for training neural networks' parameters of utterances containing OOV words (sentence-level), or rescale the gradient used for back-propagation for OOV words (word-level), when fine-tuning a previously trained model on synthetic audio. To overcome catastrophic forgetting, we also explore the combination of loss rescaling and model regularization, i.e. L2 regularization and elastic weight consolidation (EWC). Compared with previous methods that just fine-tune synthetic audio with EWC, the experimental results on the LibriSpeech benchmark reveal that our proposed loss rescaling approach can achieve significant improvement on the recall rate with only a slight decrease on word error rate. Moreover, word-level rescaling is more stable than utterance-level rescaling and leads to higher recall rates and precision on OOV word recognition. Furthermore, our proposed combined loss rescaling and weight consolidation methods can support continual learning of an ASR system., Comment: Neural Networks, Volume 161, April 2023, Pages 494-504

We study a class of reinforcement learning problems where the reward signals for policy learning are generated by an internal reward model that is dependent on and jointly optimized with the policy. This interdependence between the policy and the reward model leads to an unstable learning process because reward signals from an immature reward model are noisy and impede policy learning, and conversely, an under-optimized policy impedes reward estimation learning. We call this learning setting $\textit{Internally Rewarded Reinforcement Learning}$ (IRRL) as the reward is not provided directly by the environment but $\textit{internally}$ by a reward model. In this paper, we formally formulate IRRL and present a class of problems that belong to IRRL. We theoretically derive and empirically analyze the effect of the reward function in IRRL and based on these analyses propose the clipped linear reward function. Experimental results show that the proposed reward function can consistently stabilize the training process by reducing the impact of reward noise, which leads to faster convergence and higher performance compared with baselines in diverse tasks., Comment: Accepted at ICML 2023. Update: adopt the term "reward model" instead of using "critic" to prevent confusion with the term "critic" in actor-critic algorithms. Project webpage at https://ir-rl.github.io

Human infant learning happens during exploration of the environment, by interaction with objects, and by listening to and repeating utterances casually, which is analogous to unsupervised learning. Only occasionally, a learning infant would receive a matching verbal description of an action it is committing, which is similar to supervised learning. Such a learning mechanism can be mimicked with deep learning. We model this weakly supervised learning paradigm using our Paired Gated Autoencoders (PGAE) model, which combines an action and a language autoencoder. After observing a performance drop when reducing the proportion of supervised training, we introduce the Paired Transformed Autoencoders (PTAE) model, using Transformer-based crossmodal attention. PTAE achieves significantly higher accuracy in language-to-action and action-to-language translations, particularly in realistic but difficult cases when only few supervised training samples are available. We also test whether the trained model behaves realistically with conflicting multimodal input. In accordance with the concept of incongruence in psychology, conflict deteriorates the model output. Conflicting action input has a more severe impact than conflicting language input, and more conflicting features lead to larger interference. PTAE can be trained on mostly unlabelled data where labeled data is scarce, and it behaves plausibly when tested with incongruent input., Comment: Published in: Applied Artificial Intelligence, 37:1, 2179167

Human speech can be characterized by different components, including semantic content, speaker identity and prosodic information. Significant progress has been made in disentangling representations for semantic content and speaker identity in Automatic Speech Recognition (ASR) and speaker verification tasks respectively. However, it is still an open challenging research question to extract prosodic information because of the intrinsic association of different attributes, such as timbre and rhythm, and because of the need for supervised training schemes to achieve robust large-scale and speaker-independent ASR. The aim of this paper is to address the disentanglement of emotional prosody from speech based on unsupervised reconstruction. Specifically, we identify, design, implement and integrate three crucial components in our proposed speech reconstruction model Prosody2Vec: (1) a unit encoder that transforms speech signals into discrete units for semantic content, (2) a pretrained speaker verification model to generate speaker identity embeddings, and (3) a trainable prosody encoder to learn prosody representations. We first pretrain the Prosody2Vec representations on unlabelled emotional speech corpora, then fine-tune the model on specific datasets to perform Speech Emotion Recognition (SER) and Emotional Voice Conversion (EVC) tasks. Both objective (weighted and unweighted accuracies) and subjective (mean opinion score) evaluations on the EVC task suggest that Prosody2Vec effectively captures general prosodic features that can be smoothly transferred to other emotional speech. In addition, our SER experiments on the IEMOCAP dataset reveal that the prosody features learned by Prosody2Vec are complementary and beneficial for the performance of widely used speech pretraining models and surpass the state-of-the-art methods when combining Prosody2Vec with HuBERT representations., Comment: Accepted by IEEE/ACM Transactions on Audio, Speech, and Language Processing

The task of emotion recognition in conversations (ERC) benefits from the availability of multiple modalities, as provided, for example, in the video-based Multimodal EmotionLines Dataset (MELD). However, only a few research approaches use both acoustic and visual information from the MELD videos. There are two reasons for this: First, label-to-video alignments in MELD are noisy, making those videos an unreliable source of emotional speech data. Second, conversations can involve several people in the same scene, which requires the localisation of the utterance source. In this paper, we introduce MELD with Fixed Audiovisual Information via Realignment (MELD-FAIR) by using recent active speaker detection and automatic speech recognition models, we are able to realign the videos of MELD and capture the facial expressions from speakers in 96.92% of the utterances provided in MELD. Experiments with a self-supervised voice recognition model indicate that the realigned MELD-FAIR videos more closely match the transcribed utterances given in the MELD dataset. Finally, we devise a model for emotion recognition in conversations trained on the realigned MELD-FAIR videos, which outperforms state-of-the-art models for ERC based on vision alone. This indicates that localising the source of speaking activities is indeed effective for extracting facial expressions from the uttering speakers and that faces provide more informative visual cues than the visual features state-of-the-art models have been using so far. The MELD-FAIR realignment data, and the code of the realignment procedure and of the emotional recognition, are available at https://github.com/knowledgetechnologyuhh/MELD-FAIR., Comment: 17 pages, 8 figures, 7 tables, Published in Neurocomputing

Large-scale commonsense knowledge bases empower a broad range of AI applications, where the automatic extraction of commonsense knowledge (CKE) is a fundamental and challenging problem. CKE from text is known for suffering from the inherent sparsity and reporting bias of commonsense in text. Visual perception, on the other hand, contains rich commonsense knowledge about real-world entities, e.g., (person, can_hold, bottle), which can serve as promising sources for acquiring grounded commonsense knowledge. In this work, we present CLEVER, which formulates CKE as a distantly supervised multi-instance learning problem, where models learn to summarize commonsense relations from a bag of images about an entity pair without any human annotation on image instances. To address the problem, CLEVER leverages vision-language pre-training models for deep understanding of each image in the bag, and selects informative instances from the bag to summarize commonsense entity relations via a novel contrastive attention mechanism. Comprehensive experimental results in held-out and human evaluation show that CLEVER can extract commonsense knowledge in promising quality, outperforming pre-trained language model-based methods by 3.9 AUC and 6.4 mAUC points. The predicted commonsense scores show strong correlation with human judgment with a 0.78 Spearman coefficient. Moreover, the extracted commonsense can also be grounded into images with reasonable interpretability. The data and codes can be obtained at https://github.com/thunlp/CLEVER., Comment: Accepted by AAAI 2023

Humans can effortlessly modify various prosodic attributes, such as the placement of stress and the intensity of sentiment, to convey a specific emotion while maintaining consistent linguistic content. Motivated by this capability, we propose EmoAug, a novel style transfer model designed to enhance emotional expression and tackle the data scarcity issue in speech emotion recognition tasks. EmoAug consists of a semantic encoder and a paralinguistic encoder that represent verbal and non-verbal information respectively. Additionally, a decoder reconstructs speech signals by conditioning on the aforementioned two information flows in an unsupervised fashion. Once training is completed, EmoAug enriches expressions of emotional speech with different prosodic attributes, such as stress, rhythm and intensity, by feeding different styles into the paralinguistic encoder. EmoAug enables us to generate similar numbers of samples for each class to tackle the data imbalance issue as well. Experimental results on the IEMOCAP dataset demonstrate that EmoAug can successfully transfer different speaking styles while retaining the speaker identity and semantic content. Furthermore, we train a SER model with data augmented by EmoAug and show that the augmented model not only surpasses the state-of-the-art supervised and self-supervised methods but also overcomes overfitting problems caused by data imbalance. Some audio samples can be found on our demo website., Comment: Accepted by ICASSP2024

Sound is one of the most informative and abundant modalities in the real world while being robust to sense without contacts by small and cheap sensors that can be placed on mobile devices. Although deep learning is capable of extracting information from multiple sensory inputs, there has been little use of sound for the control and learning of robotic actions. For unsupervised reinforcement learning, an agent is expected to actively collect experiences and jointly learn representations and policies in a self-supervised way. We build realistic robotic manipulation scenarios with physics-based sound simulation and propose the Intrinsic Sound Curiosity Module (ISCM). The ISCM provides feedback to a reinforcement learner to learn robust representations and to reward a more efficient exploration behavior. We perform experiments with sound enabled during pre-training and disabled during adaptation, and show that representations learned by ISCM outperform the ones by vision-only baselines and pre-trained policies can accelerate the learning process when applied to downstream tasks., Comment: Accepted at IROS 2022

Handling various robot action-language translation tasks flexibly is an essential requirement for natural interaction between a robot and a human. Previous approaches require change in the configuration of the model architecture per task during inference, which undermines the premise of multi-task learning. In this work, we propose the paired gated autoencoders (PGAE) for flexible translation between robot actions and language descriptions in a tabletop object manipulation scenario. We train our model in an end-to-end fashion by pairing each action with appropriate descriptions that contain a signal informing about the translation direction. During inference, our model can flexibly translate from action to language and vice versa according to the given language signal. Moreover, with the option to use a pretrained language model as the language encoder, our model has the potential to recognise unseen natural language input. Another capability of our model is that it can recognise and imitate actions of another agent by utilising robot demonstrations. The experiment results highlight the flexible bidirectional translation capabilities of our approach alongside with the ability to generalise to the actions of the opposite-sitting agent., Comment: Accepted at the 31st International Conference on Artificial Neural Networks (ICANN 2022)

Spatial reasoning poses a particular challenge for intelligent agents and is at the same time a prerequisite for their successful interaction and communication in the physical world. One such reasoning task is to describe the position of a target object with respect to the intrinsic orientation of some reference object via relative directions. In this paper, we introduce GRiD-A-3D, a novel diagnostic visual question-answering (VQA) dataset based on abstract objects. Our dataset allows for a fine-grained analysis of end-to-end VQA models' capabilities to ground relative directions. At the same time, model training requires considerably fewer computational resources compared with existing datasets, yet yields a comparable or even higher performance. Along with the new dataset, we provide a thorough evaluation based on two widely known end-to-end VQA architectures trained on GRiD-A-3D. We demonstrate that within a few epochs, the subtasks required to reason over relative directions, such as recognizing and locating objects in a scene and estimating their intrinsic orientations, are learned in the order in which relative directions are intuitively processed.

Understanding spatial relations is essential for intelligent agents to act and communicate in the physical world. Relative directions are spatial relations that describe the relative positions of target objects with regard to the intrinsic orientation of reference objects. Grounding relative directions is more difficult than grounding absolute directions because it not only requires a model to detect objects in the image and to identify spatial relation based on this information, but it also needs to recognize the orientation of objects and integrate this information into the reasoning process. We investigate the challenging problem of grounding relative directions with end-to-end neural networks. To this end, we provide GRiD-3D, a novel dataset that features relative directions and complements existing visual question answering (VQA) datasets, such as CLEVR, that involve only absolute directions. We also provide baselines for the dataset with two established end-to-end VQA models. Experimental evaluations show that answering questions on relative directions is feasible when questions in the dataset simulate the necessary subtasks for grounding relative directions. We discover that those subtasks are learned in an order that reflects the steps of an intuitive pipeline for processing relative directions., Comment: Accepted to IJCAI 2022

Large datasets as required for deep learning of lip reading do not exist in many languages. In this paper we present the dataset GLips (German Lips) consisting of 250,000 publicly available videos of the faces of speakers of the Hessian Parliament, which was processed for word-level lip reading using an automatic pipeline. The format is similar to that of the English language LRW (Lip Reading in the Wild) dataset, with each video encoding one word of interest in a context of 1.16 seconds duration, which yields compatibility for studying transfer learning between both datasets. By training a deep neural network, we investigate whether lip reading has language-independent features, so that datasets of different languages can be used to improve lip reading models. We demonstrate learning from scratch and show that transfer learning from LRW to GLips and vice versa improves learning speed and performance, in particular for the validation set., Comment: Accepted to LREC 2022

Human infants learn language while interacting with their environment in which their caregivers may describe the objects and actions they perform. Similar to human infants, artificial agents can learn language while interacting with their environment. In this work, first, we present a neural model that bidirectionally binds robot actions and their language descriptions in a simple object manipulation scenario. Building on our previous Paired Variational Autoencoders (PVAE) model, we demonstrate the superiority of the variational autoencoder over standard autoencoders by experimenting with cubes of different colours, and by enabling the production of alternative vocabularies. Additional experiments show that the model's channel-separated visual feature extraction module can cope with objects of different shapes. Next, we introduce PVAE-BERT, which equips the model with a pretrained large-scale language model, i.e., Bidirectional Encoder Representations from Transformers (BERT), enabling the model to go beyond comprehending only the predefined descriptions that the network has been trained on; the recognition of action descriptions generalises to unconstrained natural language as the model becomes capable of understanding unlimited variations of the same descriptions. Our experiments suggest that using a pretrained language model as the language encoder allows our approach to scale up for real-world scenarios with instructions from human users., Comment: Published in: IEEE Transactions on Cognitive and Developmental Systems, 15:4, 3204452

The aim of this work is to investigate the impact of crossmodal self-supervised pre-training for speech reconstruction (video-to-audio) by leveraging the natural co-occurrence of audio and visual streams in videos. We propose LipSound2 which consists of an encoder-decoder architecture and location-aware attention mechanism to map face image sequences to mel-scale spectrograms directly without requiring any human annotations. The proposed LipSound2 model is firstly pre-trained on $\sim$2400h multi-lingual (e.g. English and German) audio-visual data (VoxCeleb2). To verify the generalizability of the proposed method, we then fine-tune the pre-trained model on domain-specific datasets (GRID, TCD-TIMIT) for English speech reconstruction and achieve a significant improvement on speech quality and intelligibility compared to previous approaches in speaker-dependent and -independent settings. In addition to English, we conduct Chinese speech reconstruction on the CMLR dataset to verify the impact on transferability. Lastly, we train the cascaded lip reading (video-to-text) system by fine-tuning the generated audios on a pre-trained speech recognition system and achieve state-of-the-art performance on both English and Chinese benchmark datasets., Comment: ACCEPTED IN IEEE Transactions on Neural Networks and Learning Systems

Lifelong learning is a long-standing aim for artificial agents that act in dynamic environments, in which an agent needs to accumulate knowledge incrementally without forgetting previously learned representations. We investigate methods for learning from data produced by event cameras and compare techniques to mitigate forgetting while learning incrementally. We propose a model that is composed of both, feature extraction and continuous learning. Furthermore, we introduce a habituation-based method to mitigate forgetting. Our experimental results show that the combination of different techniques can help to avoid catastrophic forgetting while learning incrementally from the features provided by the extraction module., Comment: In Proceedings of the 29th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning

The strong relation between face and voice can aid active speaker detection systems when faces are visible, even in difficult settings, when the face of a speaker is not clear or when there are several people in the same scene. By being capable of estimating the frontal facial representation of a person from his/her speech, it becomes easier to determine whether he/she is a potential candidate for being classified as an active speaker, even in challenging cases in which no mouth movement is detected from any person in that same scene. By incorporating a face-voice association neural network into an existing state-of-the-art active speaker detection model, we introduce FaVoA (Face-Voice Association Ambiguous Speaker Detector), a neural network model that can correctly classify particularly ambiguous scenarios. FaVoA not only finds positive associations, but helps to rule out non-matching face-voice associations, where a face does not match a voice. Its use of a gated-bimodal-unit architecture for the fusion of those models offers a way to quantitatively determine how much each modality contributes to the classification.

Neural networks can be powerful function approximators, which are able to model high-dimensional feature distributions from a subset of examples drawn from the target distribution. Naturally, they perform well at generalizing within the limits of their target function, but they often fail to generalize outside of the explicitly learned feature space. It is therefore an open research topic whether and how neural network-based architectures can be deployed for systematic reasoning. Many studies have shown evidence for poor generalization, but they often work with abstract data or are limited to single-channel input. Humans, however, learn and interact through a combination of multiple sensory modalities, and rarely rely on just one. To investigate compositional generalization in a multimodal setting, we generate an extensible dataset with multimodal input sequences from simulation. We investigate the influence of the underlying training data distribution on compostional generalization in a minimal LSTM-based network trained in a supervised, time continuous setting. We find compositional generalization to fail in simple setups while improving with the number of objects, actions, and particularly with a lot of color overlaps between objects. Furthermore, multimodality strongly improves compositional generalization in settings where a pure vision model struggles to generalize.

Reasoning about potential occlusions is essential for robots to efficiently predict whether an object exists in an environment. Though existing work shows that a robot with active perception can achieve various tasks, it is still unclear if occlusion reasoning can be achieved. To answer this question, we introduce the task of robotic object existence prediction: when being asked about an object, a robot needs to move as few steps as possible around a table with randomly placed objects to predict whether the queried object exists. To address this problem, we propose a novel recurrent neural network model that can be jointly trained with supervised and reinforcement learning methods using a curriculum training strategy. Experimental results show that 1) both active perception and occlusion reasoning are necessary to successfully achieve the task; 2) the proposed model demonstrates a good occlusion reasoning ability by achieving a similar prediction accuracy to an exhaustive exploration baseline while requiring only about $10\%$ of the baseline's number of movement steps on average; and 3) the model generalizes to novel object combinations with a moderate loss of accuracy., Comment: Accepted at IROS 2021

In recent years some researchers have explored the use of reinforcement learning (RL) algorithms as key components in the solution of various natural language processing tasks. For instance, some of these algorithms leveraging deep neural learning have found their way into conversational systems. This paper reviews the state of the art of RL methods for their possible use for different problems of natural language processing, focusing primarily on conversational systems, mainly due to their growing relevance. We provide detailed descriptions of the problems as well as discussions of why RL is well-suited to solve them. Also, we analyze the advantages and limitations of these methods. Finally, we elaborate on promising research directions in natural language processing that might benefit from reinforcement learning.

Scene graph generation aims to identify objects and their relations in images, providing structured image representations that can facilitate numerous applications in computer vision. However, scene graph models usually require supervised learning on large quantities of labeled data with intensive human annotation. In this work, we propose visual distant supervision, a novel paradigm of visual relation learning, which can train scene graph models without any human-labeled data. The intuition is that by aligning commonsense knowledge bases and images, we can automatically create large-scale labeled data to provide distant supervision for visual relation learning. To alleviate the noise in distantly labeled data, we further propose a framework that iteratively estimates the probabilistic relation labels and eliminates the noisy ones. Comprehensive experimental results show that our distantly supervised model outperforms strong weakly supervised and semi-supervised baselines. By further incorporating human-labeled data in a semi-supervised fashion, our model outperforms state-of-the-art fully supervised models by a large margin (e.g., 8.3 micro- and 7.8 macro-recall@50 improvements for predicate classification in Visual Genome evaluation). We make the data and code for this paper publicly available at https://github.com/thunlp/VisualDS., Comment: Accepted by ICCV 2021

In domains where computational resources and labeled data are limited, such as in robotics, deep networks with millions of weights might not be the optimal solution. In this paper, we introduce a connectivity scheme for pyramidal architectures to increase their capacity for learning features. Experiments on facial expression recognition of unseen people demonstrate that our approach is a potential candidate for applications with restricted resources, due to good generalization performance and low computational cost. We show that our approach generalizes as well as convolutional architectures in this task but uses fewer trainable parameters and is more robust for low-resolution faces.

Using a model of the environment, reinforcement learning agents can plan their future moves and achieve superhuman performance in board games like Chess, Shogi, and Go, while remaining relatively sample-efficient. As demonstrated by the MuZero Algorithm, the environment model can even be learned dynamically, generalizing the agent to many more tasks while at the same time achieving state-of-the-art performance. Notably, MuZero uses internal state representations derived from real environment states for its predictions. In this paper, we bind the model's predicted internal state representation to the environment state via two additional terms: a reconstruction model loss and a simpler consistency loss, both of which work independently and unsupervised, acting as constraints to stabilize the learning process. Our experiments show that this new integration of reconstruction model loss and simpler consistency loss provide a significant performance increase in OpenAI Gym environments. Our modifications also enable self-supervised pretraining for MuZero, so the algorithm can learn about environment dynamics before a goal is made available.

Human infants are able to acquire natural language seemingly easily at an early age. Their language learning seems to occur simultaneously with learning other cognitive functions as well as with playful interactions with the environment and caregivers. From a neuroscientific perspective, natural language is embodied, grounded in most, if not all, sensory and sensorimotor modalities, and acquired by means of crossmodal integration. However, characterising the underlying mechanisms in the brain is difficult and explaining the grounding of language in crossmodal perception and action remains challenging. In this paper, we present a neurocognitive model for language grounding which reflects bio-inspired mechanisms such as an implicit adaptation of timescales as well as end-to-end multimodal abstraction. It addresses developmental robotic interaction and extends its learning capabilities using larger-scale knowledge-based data. In our scenario, we utilise the humanoid robot NICO in obtaining the EMIL data collection, in which the cognitive robot interacts with objects in a children's playground environment while receiving linguistic labels from a caregiver. The model analysis shows that crossmodally integrated representations are sufficient for acquiring language merely from sensory input through interaction with objects in an environment. The representations self-organise hierarchically and embed temporal and spatial information through composition and decomposition. This model can also provide the basis for further crossmodal integration of perceptually grounded cognitive representations.

Target speech separation refers to isolating target speech from a multi-speaker mixture signal by conditioning on auxiliary information about the target speaker. Different from the mainstream audio-visual approaches which usually require simultaneous visual streams as additional input, e.g. the corresponding lip movement sequences, in our approach we propose the novel use of a single face profile of the target speaker to separate expected clean speech. We exploit the fact that the image of a face contains information about the person's speech sound. Compared to using a simultaneous visual sequence, a face image is easier to obtain by pre-enrollment or on websites, which enables the system to generalize to devices without cameras. To this end, we incorporate face embeddings extracted from a pretrained model for face recognition into the speech separation, which guide the system in predicting a target speaker mask in the time-frequency domain. The experimental results show that a pre-enrolled face image is able to benefit separating expected speech signals. Additionally, face information is complementary to voice reference and we show that further improvement can be achieved when combing both face and voice embeddings.

Combining model-based and model-free learning systems has been shown to improve the sample efficiency of learning to perform complex robotic tasks. However, dual-system approaches fail to consider the reliability of the learned model when it is applied to make multiple-step predictions, resulting in a compounding of prediction errors and performance degradation. In this paper, we present a novel dual-system motor learning approach where a meta-controller arbitrates online between model-based and model-free decisions based on an estimate of the local reliability of the learned model. The reliability estimate is used in computing an intrinsic feedback signal, encouraging actions that lead to data that improves the model. Our approach also integrates arbitration with imagination where a learned latent-space model generates imagined experiences, based on its local reliability, to be used as additional training data. We evaluate our approach against baseline and state-of-the-art methods on learning vision-based robotic grasping in simulation and real world. The results show that our approach outperforms the compared methods and learns near-optimal grasping policies in dense- and sparse-reward environments.

The recognition of emotion and dialogue acts enriches conversational analysis and help to build natural dialogue systems. Emotion interpretation makes us understand feelings and dialogue acts reflect the intentions and performative functions in the utterances. However, most of the textual and multi-modal conversational emotion corpora contain only emotion labels but not dialogue acts. To address this problem, we propose to use a pool of various recurrent neural models trained on a dialogue act corpus, with and without context. These neural models annotate the emotion corpora with dialogue act labels, and an ensemble annotator extracts the final dialogue act label. We annotated two accessible multi-modal emotion corpora: IEMOCAP and MELD. We analyzed the co-occurrence of emotion and dialogue act labels and discovered specific relations. For example, Accept/Agree dialogue acts often occur with the Joy emotion, Apology with Sadness, and Thanking with Joy. We make the Emotional Dialogue Acts (EDA) corpus publicly available to the research community for further study and analysis., Comment: Proceeding of the LREC 2020

Establishing associations between the structure and the generalisation ability of deep neural networks (DNNs) is a challenging task in modern machine learning. Producing solutions to this challenge will bring progress both in the theoretical understanding of DNNs and in building new architectures efficiently. In this work, we address this challenge by developing a new complexity measure based on the concept of {Periodic Spectral Ergodicity} (PSE) originating from quantum statistical mechanics. Based on this measure a technique is devised to quantify the complexity of deep neural networks from the learned weights and traversing the network connectivity in a sequential manner, hence the term cascading PSE (cPSE), as an empirical complexity measure. This measure will capture both topological and internal neural processing complexity simultaneously. Because of this cascading approach, i.e., a symmetric divergence of PSE on the consecutive layers, it is possible to use this measure for Neural Architecture Search (NAS). We demonstrate the usefulness of this measure in practice on two sets of vision models, ResNet and VGG, and sketch the computation of cPSE for more complex network structures., Comment: 9 pages, 5 figures

Combining model-based and model-free deep reinforcement learning has shown great promise for improving sample efficiency on complex control tasks while still retaining high performance. Incorporating imagination is a recent effort in this direction inspired by human mental simulation of motor behavior. We propose a learning-adaptive imagination approach which, unlike previous approaches, takes into account the reliability of the learned dynamics model used for imagining the future. Our approach learns an ensemble of disjoint local dynamics models in latent space and derives an intrinsic reward based on learning progress, motivating the controller to take actions leading to data that improves the models. The learned models are used to generate imagined experiences, augmenting the training set of real experiences. We evaluate our approach on learning vision-based robotic grasping and show that it significantly improves sample efficiency and achieves near-optimal performance in a sparse reward environment., Comment: In: Proceedings of the Joint IEEE International Conference on Development and Learning and on Epigenetic Robotics (ICDL-EpiRob), Oslo, Norway, Aug. 19-22, 2019

Selective attention plays an essential role in information acquisition and utilization from the environment. In the past 50 years, research on selective attention has been a central topic in cognitive science. Compared with unimodal studies, crossmodal studies are more complex but necessary to solve real-world challenges in both human experiments and computational modeling. Although an increasing number of findings on crossmodal selective attention have shed light on humans' behavioral patterns and neural underpinnings, a much better understanding is still necessary to yield the same benefit for computational intelligent agents. This article reviews studies of selective attention in unimodal visual and auditory and crossmodal audiovisual setups from the multidisciplinary perspectives of psychology and cognitive neuroscience, and evaluates different ways to simulate analogous mechanisms in computational models and robotics. We discuss the gaps between these fields in this interdisciplinary review and provide insights about how to use psychological findings and theories in artificial intelligence from different perspectives., Comment: 29pages, 5 figures, 1 table, journal article

Recent success in deep reinforcement learning for continuous control has been dominated by model-free approaches which, unlike model-based approaches, do not suffer from representational limitations in making assumptions about the world dynamics and model errors inevitable in complex domains. However, they require a lot of experiences compared to model-based approaches that are typically more sample-efficient. We propose to combine the benefits of the two approaches by presenting an integrated approach called Curious Meta-Controller. Our approach alternates adaptively between model-based and model-free control using a curiosity feedback based on the learning progress of a neural model of the dynamics in a learned latent space. We demonstrate that our approach can significantly improve the sample efficiency and achieve near-optimal performance on learning robotic reaching and grasping tasks from raw-pixel input in both dense and sparse reward settings., Comment: Accepted at IJCNN 2019

In this paper, we describe KT-Speech-Crawler: an approach for automatic dataset construction for speech recognition by crawling YouTube videos. We outline several filtering and post-processing steps, which extract samples that can be used for training end-to-end neural speech recognition systems. In our experiments, we demonstrate that a single-core version of the crawler can obtain around 150 hours of transcribed speech within a day, containing an estimated 3.5% word error rate in the transcriptions. Automatically collected samples contain reading and spontaneous speech recorded in various conditions including background noise and music, distant microphone recordings, and a variety of accents and reverberation. When training a deep neural network on speech recognition, we observed around 40\% word error rate reduction on the Wall Street Journal dataset by integrating 200 hours of the collected samples into the training set. The demo (http://emnlp-demo.lakomkin.me/) and the crawler code (https://github.com/EgorLakomkin/KTSpeechCrawler) are publicly available., Comment: Accepted at the Conference on Empirical Methods in Natural Language Processing 2018, Brussels, Belgium

Previous work on emotion recognition demonstrated a synergistic effect of combining several modalities such as auditory, visual, and transcribed text to estimate the affective state of a speaker. Among these, the linguistic modality is crucial for the evaluation of an expressed emotion. However, manually transcribed spoken text cannot be given as input to a system practically. We argue that using ground-truth transcriptions during training and evaluation phases leads to a significant discrepancy in performance compared to real-world conditions, as the spoken text has to be recognized on the fly and can contain speech recognition mistakes. In this paper, we propose a method of integrating an automatic speech recognition (ASR) output with a character-level recurrent neural network for sentiment recognition. In addition, we conduct several experiments investigating sentiment recognition for human-robot interaction in a noise-realistic scenario which is challenging for the ASR systems. We quantify the improvement compared to using only the acoustic modality in sentiment recognition. We demonstrate the effectiveness of this approach on the Multimodal Corpus of Sentiment Intensity (MOSI) by achieving 73,6% accuracy in a binary sentiment classification task, exceeding previously reported results that use only acoustic input. In addition, we set a new state-of-the-art performance on the MOSI dataset (80.4% accuracy, 2% absolute improvement)., Comment: Accepted at the 2019 International Conference on Robotics and Automation (ICRA) will be held on May 20-24, 2019 in Montreal, Canada

In this paper, we present a new intrinsically motivated actor-critic algorithm for learning continuous motor skills directly from raw visual input. Our neural architecture is composed of a critic and an actor network. Both networks receive the hidden representation of a deep convolutional autoencoder which is trained to reconstruct the visual input, while the centre-most hidden representation is also optimized to estimate the state value. Separately, an ensemble of predictive world models generates, based on its learning progress, an intrinsic reward signal which is combined with the extrinsic reward to guide the exploration of the actor-critic learner. Our approach is more data-efficient and inherently more stable than the existing actor-critic methods for continuous control from pixel data. We evaluate our algorithm for the task of learning robotic reaching and grasping skills on a realistic physics simulator and on a humanoid robot. The results show that the control policies learned with our approach can achieve better performance than the compared state-of-the-art and baseline algorithms in both dense-reward and challenging sparse-reward settings.

Spoken language understanding is one of the key factors in a dialogue system, and a context in a conversation plays an important role to understand the current utterance. In this work, we demonstrate the importance of context within the dialogue for neural network models through an online web interface live demo. We developed two different neural models: a model that does not use context and a context-based model. The no-context model classifies dialogue acts at an utterance-level whereas the context-based model takes some preceding utterances into account. We make these trained neural models available as a live demo called Discourse-Wizard using a modular server architecture. The live demo provides an easy to use interface for conversational analysis and for discovering deep discourse structures in a conversation., Comment: Submitted to EMNLP 2018: System Demonstrations