3.1 Background

In recent years, Artificial Intelligence (AI) has achieved important landmarks in matching or surpassing human level performance on a number of high level tasks (playing chess and go, driving cars, categorizing picture, etc., 30, 34, 39, 29, 36). These strong advances were obtained by deploying on large amounts of data, massively parallel learning architectures with simple brain-inspired ‘neuronal’ elements. However, humans brains still outperform machines in several key areas (language, social interactions, common sense reasoning, motor skills), and are more flexible : Whereas machines require extensive expert knowledge and massive training for each particular application, humans learn autonomously over several time scales: over the developmental scale (months), humans infants acquire cognitive skills with noisy data and little or no expert feedback (weakly/unsupervised learning)1; over the short time scale (minutes, seconds), humans combine previously acquired skills to solve new tasks and apply rules systematically to draw inferences on the basis of extremely scarce data (learning to learn, domain adaptation, one- or zero-shot learning) 32.

The general aim of CoML, following the roadmap described in 1, is to bridge the gap in cognitive flexibility between humans and machines learning in language processing and common sense reasoning by reverse engineering how young children between 1 and 4 years of age learn from their environment. We conduct work along two axes: the first one, which we called Developmental AI is focused on building infant inspired machine learning algorithms. The second axis is devoted to using the developed algorithms to conduct quantitative studies of how infant learn across diverse environments.

3.2 Weakly/Unsupervised Learning

Much of standard machine learning is construed as regression or classification problems (mapping input data to expert-provided labels). Human infants rarely learn in this fashion, at least before going to school: they learn language, social cognition, and common sense autonomously (without expert labels) and when adults provide feedback, it is ambiguous and noisy and cannot be taken as a gold standard. Modeling or mimicking such achievement requires deploying unsupervised or weakly supervised algorithms which are less well known than their supervised counterparts.

We take inspiration from infant’s landmarks during their first years of life: they are able to learn acoustic models, a lexicon, and susbtantive elements of language models and world models from raw sensory inputs. Building on previous work 3, 7, 11, we use DNN and Bayesian architectures to model the emergence of linguistic representations without supervision. Our focus is to establish how the labels in supervised settings can be replaced by weaker signals coming either from multi-modal input or from hierarchically organised linguistic levels.

At the level of phonetic representations, we study how cross-modal information (lips and self feedback from articulation) can supplement top-down lexical information in a weakly supervised setting. We use Siamese architectures or Deep CCA algorithms to combine the different views. We study how an attentional framework and uncertainty estimation can flexibly combine these informations in order to adapt to situations where one view is selectively degraded.

At the level of lexical representations, we study how audio/visual parallel information (ie. descriptions of images or activities) can help in segmenting and clustering word forms, and vice versa, help in deriving useful visual features. To achieve this, we will use architectures deployed in image captioning or sequence to sequence translation 37.

At the level of semantic and conceptual representations, we study how it is possible to learn elements of the laws of physics through the observation of videos (object permanence, solidity, spatio-temporal continuity, inertia, etc.), and how objects and relations between objects are mapped onto language.

3.3 Evaluating Machine Intelligence

Increasingly, complicated machine learning systems are being incorporated into real-life applications (e.g. self-driving cars, personal assistants), even though they cannot be formally verified, guaranteed statistically, nor even explained. In these cases, a well defined empirical approach to evaluation can offer interesting insights into the functioning and offer some control over these algorithms.

Several approaches exist to evaluate the 'cognitive' abilities of machines, from the subjective comparison of human and machine performance 38 to application-specific metrics (e.g., in speech, word error rate). A recent idea consist in evaluating an AI system in terms of it's abilities31 , i.e., functional components within a more global cognitive architecture 35. Psychophysical testing can offer batteries of tests using simple tasks that are easy to understand by humans or animals (e.g, judging whether two stimuli are same or different, or judging whether one stimulus is ‘typical’) which can be made selective to a specific component and to rare but difficult or adversarial cases. Evaluations of learning rate, domain adaptation and transfer learning are simple applications of these measures. Psychophysically inspired tests have been proposed for unsupervised speech and language learning 10, 33.

3.4 Documenting human learning

Infants learn their first language in a spontaneous fashion, across a lot of variation in amount of speech and the nature of the infant/adult interaction. In some linguistic communities, adults barely address infants until they can themselves speak. Despite these large variations in quantity and content, language learning proceeds at similar paces. Documenting such resilience is an essential step in understanding the nature of the learning algorithms used by human infants. Hence, we propose to collect and/or analyse large datasets of inputs to infants and correlate this with outcome measure (phonetic learning, vocabulary growth, syntactic learning, etc.).

4.1 Speech processing for underresourced languages

We plan to apply our algorithms for the unsupervised discovery of speech units to problems relevant to language documentation and the construction of speech processing pipelines for underresourced languages.

4.2 Tools for the analysis of naturalistic speech corpora

Daylong recordings of speech in the wild gives rise a to number of specific analysis difficulties. We plan to use our expertise in speech processing to develop tools for performing signal processing and helping annotation of such resources for the purpose of phonetic or linguistic analysis.

5.1 Footprint of research activities

The footprint of the CoML team due to travel was close to zero since, because of the sanitary conditions, all conferences were attended via video conference. The compute footprint was that of our 4 GPU cluster, used on average 30% of the time, to which we should add the compute of our accounts at Jean Zay (we could not access the data at the time of the report).

5.2 Impact of research results

Our fundamental work in unsupervised learning algorithms are very early stage and have up to now no known environmental/societal application. Our applicative work is dedicated to develop spoken language annotation and analysis tools for researchers, which should help conduct research in clinical and developmental areas (Health and Well Being, and early Education).

7.1 New software

8.1 Unsupervised learning

Humans learn to speak and to perceive the world in a largely self-supervised fashion. Yet, most of machine learning is still devoted to supervised algorithms that rely on abundant quantities of human labelled data. We have used humans as sources of inspiration for developing novel machine learning algorithms in order to push the field towards self-supervised learning.

• In 13, we introduced Generative Spoken Language Modeling, the task of learning the acoustic and linguistic characteristics of a language from raw audio (no text, no labels), and a set of metrics to automatically evaluate the learned representations at acoustic and linguistic levels for both encoding and generation. We set up baseline systems consisting of a discrete speech encoder (returning pseudo-text units), a generative language model (trained on pseudo-text), and a speech decoder (generating a waveform from pseudo-text) all trained without supervision and validate the proposed metrics with human evaluation. Across 3 speech encoders (CPC, wav2vec 2.0, HuBERT), we find that the number of discrete units (50, 100, or 200) matters in a task-dependent and encoder-dependent way, and that some combinations approach text-based systems.
• In 20, we proposed using self-supervised discrete representations for the task of speech resynthesis. To generate disentangled representation, we separately extract low-bitrate representations for speech content, prosodic information, and speaker identity. This allows to synthesize speech in a controllable manner. We analyze various state-of-the-art, self-supervised representation learning methods and shed light on the advantages of each method while considering reconstruction quality and disentanglement properties. Specifically, we evaluate the F0 reconstruction, speaker identification performance (for both resynthesis and voice conversion), recordings' intelligibility, and overall quality using subjective human evaluation. Lastly, we demonstrate how these representations can be used for an ultra-lightweight speech codec. Using the obtained representations, we can get to a rate of 365 bits per second while providing better speech quality than the baseline methods. Audio samples can be found under the following link: speechbot.github.io/resynthesis.
• To reach human performance on complex tasks, a key ability for artificial systems is to understand physical interactions between objects, and predict future outcomes of a situation. This ability, often referred to as intuitive physics, has recently received attention and several methods were pro-posed to learn these physical rules from video sequences. Yet, most of these methods are restricted to the case where no, or only limited, occlusions occur. In 26, we propose a probabilistic formulation of learning intuitive physics in 3D scenes with significant inter-object occlusions. In our formulation, object positions are modeled as latent variables enabling the reconstruction of the scene. We then propose a series of approximations that make this problem tractable. Object proposals are linked across frames using a combination of a recurrent interaction network,modeling the physics in object space, and a compositional renderer, modeling the way in which objects project onto pixel space. We demonstrate significant improvements over state-of-the-art in the intuitive physics benchmark of Riochet et al.(2018). We apply our method to a second dataset with increasing levels of occlusions, showing it realistically predicts segmentation masks up to 30frames in the future. Finally, we also show results on predicting motion of objects in real videos

8.2 Datasets and Benchmarks

Self-supervised learning is a relatively new field of research. The CoML team contributes to the research by building benchmarks and organizing challenges to enable comparison between systems on a single set of metrics and cumulative progress across laboratories. The specificity of our approach is that we base our metrics on human psycholinguistics and psychophysics, enablig direct human - machine comparisons.

• In 19, we present the Zero Resource Speech Challenge 2021, which asks participants to learn a language model directly from audio, without any text or labels. The challenge is based on the Libri-light dataset, which provides up to 60k hours of audio from English audio books without any associated text. We provide a pipeline baseline system consisting on an encoder based on contrastive predictive coding (CPC), a quantizer (k-means) and a standard language model (BERT or LSTM). The metrics evaluate the learned representations at the acoustic (ABX discrimination), lexical (spot-the-word), syntactic (acceptability judgment) and semantic levels (similarity judgment). We present an overview of the eight submitted systems from four groups and discuss the main results.
• In 23, we introduce VoxPopuli, a large-scale multilingual corpus providing 400K hours of unlabeled speech data in 23 languages. It is the largest open data to date for unsupervised representation learning as well as semisupervised learning. VoxPopuli also contains 1.8K hours of transcribed speeches in 15 languages and their aligned oral interpretations into 15 target languages totaling 17.3K hours. We provide speech recognition (ASR) baselines and validate the versatility of VoxPopuli unlabeled data in semisupervised ASR and speech-to-text translation under challenging out-of-domain settings.
• In order to reach human performance on complex visual tasks, artificial systems need to incorporate a significant amount of understanding of the world in terms of macroscopic objects, movements, forces, etc. Inspired by work on intuitive physics in infants, we propose in 16 an evaluation framework which diagnoses how much a given system understands about physics by testing whether it can tell apart well matched videos of possible versus impossible events. The test requires systems to compute a physical plausibility score over an entire video. It is free of bias and can test a range of specific physical reasoning skills. We then describe the first release of a benchmark dataset aimed at learning intuitive physics in an unsupervised way, using videos constructed with a game engine. We describe two Deep Neural Network baseline systems trained with a future frame prediction objective and tested on the possible versus impossible discrimination task. The analysis of their results compared to human data gives novel insights in the potentials and limitations of next frame prediction architectures.

8.3 Language emergence in communicative agents

In this research topic, which was the focus of Rahma Chaabouni's PhD thesis 24, which was taken up by the MSR funded PhD of Mathieu Rita, we study the inductive biases of neural systems by presenting them with few or no data.

• Words categorize the semantic fields they refer to in ways that maximize communication accuracy while minimizing complexity. Focusing on the well-studied color domain, we show that artificial neural networks trained with deep-learning techniques to play a discrimination game develop communication systems whose distribution on the accuracy/complexity plane closely matches that of human languages. The observed variation among emergent color-naming systems is explained by different degrees of discriminative need, of the sort that might also characterize different human communities. Like human languages, emergent systems show a preference for relatively low-complexity solutions, even at the cost of imperfect communication. We demonstrate next that the nature of the emergent systems crucially depends on communication being discrete (as is human word usage). When continuous message passing is allowed, emergent systems become more complex and eventually less efficient. Our study suggests that efficient semantic categorization is a general property of discrete communication systems, not limited to human language. It suggests moreover that it is exactly the discrete nature of such systems that, acting as a bottleneck, pushes them toward low complexity and optimal efficiency.

8.4 Evaluation of AI algorithms

Machine learning algorithms are typically evaluated in terms of end-to-end tasks, but it is very often difficult to get a grasp of how they achieve these tasks, what could be their break point, and more generally, how they would compare to the algorithms used by humans to do the same tasks. This is especially true of Deep Learning systems which are particularly opaque. The team develops evaluation/interpretation methods based on psycholinguistic/linguistic/neuroscience criteria, and deploy them for systematic comparison of systems.

• Deep Learning models have become potential candidates for auditory neuroscience research, thanks to their recent successes on a variety of auditory tasks. Yet, these models often lack interpretability to fully understand the exact computations that have been performed. In 15, we propose a parametrized neural network layer, that computes specific spectro-temporal modulations based on Gabor kernels (Learnable STRFs) and that is fully interpretable. We evaluated the predictive capabilities of this layer on Speech Activity Detection, Speaker Verification, Urban Sound Classification and Zebra Finch Call Type Classification. We found out that models based on this learnable parametrized neural network are on par for all tasks with the different toplines, and obtain the best performance for Speech Activity Detection. As this layer is fully interpretable, we used quantitative measures to describe the distribution of the learned spectro-temporal modulations. The filters adapted to each task and focused mostly modulation on low temporal and spectral modulations. The analyses show that the filters learned on human speech have similar spectro-temporal parameters as the ones measured directly in the human auditory cortex. Finally, equipped with the Sinkhorn distance to com- pare the learned STRFs distributions, we observed that the tasks organized in a meaningful way: the human vocalizations tasks closer to each other and bird vocalizations far away from human vocalizations and urban sounds tasks.

8.5 Simulation of language learning in infants

Supervised learning algorithms provide very interesting quantitative models of the early phases of language acquisition. When fed with realistic input, they can generate predictions that can be compared with available developmental behavioral data.

• Before they even speak, infants become attuned to the sounds of the language(s) they hear, processing native phonetic contrasts more easily than non-native ones. For example, between 6-8 months and 10-12 months, infants learning American English get better at distinguishing English [r] and [l], as in ‘rock’ vs ‘lock’, relative to infants learning Japanese. Influential accounts of this early phonetic learning phenomenon initially proposed that infants group sounds into native vowel- and consonant-like phonetic categories—like [r] and [l] in English—through a statistical clustering mechanism dubbed ‘distributional learning’. The feasibility of this mechanism for learning phonetic categories has been challenged, however. In 22, we demonstrate that a distributional learning algorithm operating on naturalistic speech can predict early phonetic learning as observed in Japanese and American English infants, suggesting that infants might learn through distributional learning after all. We further show, however, that contrary to the original distributional learning proposal, our model learns units too brief and too fine-grained acoustically to correspond to phonetic categories. This challenges the influential idea that what infants learn are phonetic categories. More broadly, our work introduces a novel mechanism-driven approach to the study of early phonetic learning, together with a quantitative modeling framework that can handle realistic input. This allows, for the first time, accounts of early phonetic learning to be linked to concrete, systematic predictions regarding infants’ attunement.
• Theories and data on language acquisition suggest a range of cues are used, ranging from information on structure found in the linguistic signal itself, to information gleaned from the environmental context or through social interaction. In 17, we propose a blueprint for computational models of the early language learner (SCALa, for Socio-Computational Architecture of Language Acquisition) that makes explicit the connection between the kinds of information available to the social learner and the computational mechanisms required to extract language-relevant information and learn from it. SCALa integrates a range of views on language acquisition, further allowing us to make precise recommendations for future large-scale empirical research.

8.6 Quantitative studies of human learning and processing

Evidently, infants are acquiring their language based on whatever linguistic input is available around them. The extent of variation that can be found across languages, cultures and socio-economic background provides strong constraints (lower bounds on data, higher bounds on noise, and variation and ambiguity) for language learning algorithms. Vice-versa, aging adults, or patients with neurological impairements show degradation in speech and language patterns which can be used to diagnose or predict the progress of the impairement.

• Nous présentons une implémentation libre (open-source) en Python de la mesure Gamma (γ) pour l'accord inter/intra-annotateurs.
• A prominent hypothesis holds that by speaking to infants in infant-directed speech (IDS) as opposed to adult-directed speech (ADS), parents help them learn phonetic categories. Specifically, two characteristics of IDS have been claimed to facilitate learning: hyperarticulation, which makes the categories more separable and variability, which makes the generalization more robust. Here, we test the separability and robustness of vowel category learning on acoustic representations of speech uttered by Japanese adults in either ADS, IDS (addressed to 18-24 month olds) or read speech (RS). Separability is determined by means of a distance measure computed between the five short vowel categories of Japanese, while robustness is assessed by testing the ability of six different machine learning algorithms trained to classify vowels to generalize on stimuli spoken by a novel speaker in ADS. Using two different speech representations, we find that hyperarticulated speech, in the case of RS, can yield better separability, and that increased between-speaker variability in ADS, can yield, for some algorithms, more robust categories. However, these conclusions do not apply to IDS, which turned out to yield neither more separable nor more robust categories compared to ADS inputs. We discuss the usefulness of machine learning algorithms run on real data to test hypotheses about the functional role of IDS.

8.7 Test of the psychological validity of AI algorithms.

In this section, we focus on the utilisation of machine learning algorithms of speech and language processing to derive testable quantitative predictions in humans (adults or infants).

8.8 Applications and tools for researchers

Some of CoMLs' activity is to produce speech and language technology tools that facilitate research into language development or clinical applications.

8.9 Interactive AI

• Curiosity is a vital metacognitive skill in educational contexts, leading to creativity, and a love of learning. And while many school systems increasingly undercut curiosity by teaching to the test, teachers are increasingly interested in how to evoke curiosity in their students to prepare them for a world in which lifelong learning and reskilling will be more and more important. One aspect of curiosity that has received little attention, however, is the role of peers in eliciting curiosity. We present what we believe to be the first theoretical framework that articulates an integrated socio-cognitive account of curiosity that ties observable behaviors in peers to underlying curiosity states. We make a bipartite distinction between individual and interpersonal functions that contribute to curiosity, and multimodal behaviors that fulfill these functions. We validate the proposed framework by leveraging a longitudinal latent variable modeling approach. Findings confirm a positive predictive relationship between the latent variables of individual and interpersonal functions and curiosity, with the interpersonal functions exercising a comparatively stronger influence. Prominent behavioral realizations of these functions are also discovered in a data-driven manner. We instantiate the proposed theoretical framework in a set of strategies and tactics that can be incorporated into learning technologies to indicate, evoke, and scaffold curiosity. This work is a step towards designing learning technologies that can recognize and evoke moment-by-moment curiosity during learning in social contexts and towards a more complete multimodal learning analytics. The underlying rationale is applicable more generally for developing computer support for other metacognitive and socio-emotional skills.
• Interaction takes place not only on the propositional level but also on the social level. In this paper, we consider rapport as an important social phenomenon in interaction. Motivated by data from the tutoring domain, we hypothesize that (i) off-task episodes are triggered by a low level of rapport and (ii) such episodes are means of raising the level of rapport. We sketch a planning model that allows off-task episodes to be triggered by (low) rapport level, which we apply to two simple examples.

10.1 International research visitors

10.2 National initiatives

10.3 Regional initiatives

• SESAME (Région Ile de France) (2021-2031; coordinating PI: E. Dupoux, 400k€). The echolalia platform. Digital platform for annotating audio/video language data and analyzing them with AI systems. Applications: language development, sign language, language pathology.

11.1 Promoting scientific activities

11.2 Teaching - Supervision - Juries

11.3 Popularization

12.1 Major publications

• 1 articleE.Emmanuel Dupoux. Cognitive Science in the era of Artificial Intelligence: A roadmap for reverse-engineering the infant language-learner.Cognition2018
  back to textback to text
• 2 inproceedingsA.Abdellah Fourtassi and E.Emmanuel Dupoux. A Rudimentary Lexicon and Semantics Help Bootstrap Phoneme Acquisition.Proceedings of the 18th Conference on Computational Natural Language Learning (CoNLL)Baltimore, Maryland USAAssociation for Computational LinguisticsJune 2014, 191-200
  DOI
• 3 inproceedingsA.Abdellah Fourtassi, T.Thomas Schatz, B.Balakrishnan Varadarajan and E.Emmanuel Dupoux. Exploring the Relative Role of Bottom-up and Top-down Information in Phoneme Learning.Proceedings of the 52nd Annual meeting of the ACL2ACLBaltimore, MarylandAssociation for Computational Linguistics2014, 1-6
  DOIback to text
• 4 inproceedingsY.Yedid Hoshen, R. J.Ron J Weiss and K. W.Kevin W Wilson. Speech acoustic modeling from raw multichannel waveforms.Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference onIEEE2015, 4624--4628
• 5 articleT.Tal Linzen, E.Emmanuel Dupoux and Y.Yoav Goldberg. Assessing the ability of LSTMs to learn syntax-sensitive dependencies.Transactions of the Association for Computational Linguistics42016, 521-535
• 6 inproceedingsT.Tal Linzen, E.Emmanuel Dupoux and B.Benjamin Spector. Quantificational features in distributional word representations.Proceedings of the Fifth Joint Conference on Lexical and Computational Semantics2016, 1-11
  DOI
• 7 articleA.Andrew Martin, S.Sharon Peperkamp and E.Emmanuel Dupoux. Learning Phonemes with a Proto-lexicon.Cognitive Science37# NGRAM # first draft written by Rozenn; new data needs to be added by DPX # totally redone by Andy with hot new data; should be submitted to Cognition in Oct # Andy sent it to Cognition on May 2009 #rejected on July 31 - We asked for the authorization of sending a full article; accepted August 10. # need 3 days of work to rewrite. # Andy has done a lot of new simulations: can now be resent as full article # andy redid the graphs; dpx and sharon rewrote the intro; Sharon does last changes Nov 18 # andy has to write the cover & responses to reviewers; sent on Dec 8,2009 # rejected May 2010- resent to JML May 25 -- rejected Dec 2010 ## dpx rewrote the intro and Andy sent a lettre de ralage (Jan 1), ## the Editor says yes but very reluctantly--submitted Feb 14 ## Rejected May 15; resent to Dev Science May20; rejected; resent to Cog Science ## Oct 14, 2011: dpx rewrote the intro completely Sent to Dev Science first submission Oct 6, 2011 Martin_PD_2012_learning_phonemnes_with_pseudolexicon.pdf2013, 103-124
  DOIback to text
• 8 articleS.Soroush Mehri, K.Kundan Kumar, I.Ishaan Gulrajani, R.Rithesh Kumar, S.Shubham Jain, J.Jose Sotelo, A.Aaron Courville and Y.Yoshua Bengio. SampleRNN: An unconditional end-to-end neural audio generation model.arXiv preprint arXiv:1612.078372016
• 9 inproceedingsT. N.Tara N Sainath, R. J.Ron J Weiss, A.Andrew Senior, K. W.Kevin W Wilson and O.Oriol Vinyals. Learning the speech front-end with raw waveform CLDNNs.Sixteenth Annual Conference of the International Speech Communication Association2015
• 10 inproceedingsT.Thomas Schatz, V.Vijayaditya Peddinti, F.Francis Bach, A.Aren Jansen, H.Hermansky Hynek and E.Emmanuel Dupoux. Evaluating speech features with the Minimal-Pair ABX task: Analysis of the classical MFC/PLP pipeline.INTERSPEECH-2013International Speech Communication AssociationLyon, France2013, 1781-1785
  back to text
• 11 inproceedingsR.Roland Thiollière, E.Ewan Dunbar, G.Gabriel Synnaeve, M.Maarten Versteegh and E.Emmanuel Dupoux. A Hybrid Dynamic Time Warping-Deep Neural Network Architecture for Unsupervised Acoustic Modeling.INTERSPEECH-20152015, 3179-3183
  back to text
• 12 articleA.Aäron Van Den Oord, S.Sander Dieleman, H.Heiga Zen, K.Karen Simonyan, O.Oriol Vinyals, A.Alex Graves, N.Nal Kalchbrenner, A.Andrew Senior and K.Koray Kavukcuoglu. Wavenet: A generative model for raw audio.CoRR abs/1609.034992016

12.2 Publications of the year

12.3 Cited publications

• 29 articleD. A.David A Ferrucci. Introduction to ``this is watson''.IBM Journal of Research and Development563.42012, 1--1
  back to text
• 30 inproceedingsK.Kaiming He, X.Xiangyu Zhang, S.Shaoqing Ren and J.Jian Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification.Proceedings of the IEEE International Conference on Computer Vision2015, 1026--1034
  back to text
• 31 articleJ.José Hernández-Orallo, F.Fernando Martínez-Plumed, U.Ute Schmid, M.Michael Siebers and D. L.David L Dowe. Computer models solving intelligence test problems: Progress and implications.Artificial Intelligence2302016, 74--107
  back to text
• 32 articleB. M.Brenden M Lake, T. D.Tomer D Ullman, J. B.Joshua B Tenenbaum and S. J.Samuel J Gershman. Building machines that learn and think like people.arXiv preprint arXiv:1604.002892016
  back to text
• 33 articleT.Tal Linzen, E.Emmanuel Dupoux and Y.Yoav Goldberg. Assessing the Ability of LSTMs to Learn Syntax-Sensitive Dependencies.Transactions of the Association for Computational Linguistics42016, 521--535
  back to text
• 34 articleC.Chaochao Lu and X.Xiaoou Tang. Surpassing human-level face verification performance on LFW with GaussianFace.arXiv preprint arXiv:1404.38402014
  back to text
• 35 articleS. T.Shane T Mueller. A partial implementation of the BICA cognitive decathlon using the Psychology Experiment Building Language (PEBL).International Journal of Machine Consciousness2022010, 273--288
  back to text
• 36 articleD.David Silver, A.Aja Huang, C. J.Chris J. Maddison, A.Arthur Guez, L.Laurent Sifre, G.George van den Driessche, J.Julian Schrittwieser, I.Ioannis Antonoglou, V.Veda Panneershelvam, M.Marc Lanctot, S.Sander Dieleman, D.Dominik Grewe, J.John Nham, N.Nal Kalchbrenner, I.Ilya Sutskever, T.Timothy Lillicrap, M.Madeleine Leach, K.Koray Kavukcuoglu, T.Thore Graepel and D.Demis Hassabis. Mastering the game of Go with deep neural networks and tree search.Nature52975872016, 484--489
  back to text
• 37 inproceedingsI.Ilya Sutskever, O.Oriol Vinyals and Q. V.Quoc V Le. Sequence to sequence learning with neural networks.Advances in neural information processing systems2014, 3104--3112
  back to text
• 38 articleA. M.Alan M. Turing. Computing machinery and intelligence.Mind592361950, 433--460
  back to text
• 39 articleW.Wayne Xiong, J.Jasha Droppo, X.Xuedong Huang, F.Frank Seide, M.Mike Seltzer, A.Andreas Stolcke, D.Dong Yu and G.Geoffrey Zweig. Achieving human parity in conversational speech recognition.arXiv preprint arXiv:1610.052562016
  back to text