J. Weston - Google

J. Weston
Are you J. Weston?

Claim your profile, edit publications, add additional information:

Contact Details

Name
J. Weston
Affiliation
Google
City
Washington
Country
United States

Pubs By Year

External Links

Pub Categories

 
Computer Science - Computation and Language (19)
 
Solar and Stellar Astrophysics (10)
 
High Energy Astrophysical Phenomena (10)
 
Computer Science - Learning (10)
 
Physics - Mesoscopic Systems and Quantum Hall Effect (8)
 
Computer Science - Artificial Intelligence (7)
 
Statistics - Machine Learning (4)
 
Computer Science - Information Retrieval (4)
 
High Energy Physics - Experiment (3)
 
Physics - Instrumentation and Detectors (3)
 
Astrophysics of Galaxies (1)
 
Computer Science - Networking and Internet Architecture (1)
 
Physics - Superconductivity (1)
 
Computer Science - Performance (1)
 
Computer Science - Neural and Evolutionary Computing (1)

Publications Authored By J. Weston

We introduce ParlAI (pronounced "par-lay"), an open-source software platform for dialog research implemented in Python, available at http://parl.ai. Its goal is to provide a unified framework for training and testing of dialog models, including multitask training, and integration of Amazon Mechanical Turk for data collection, human evaluation, and online/reinforcement learning. Read More

This paper proposes to tackle open- domain question answering using Wikipedia as the unique knowledge source: the answer to any factoid question is a text span in a Wikipedia article. This task of machine reading at scale combines the challenges of document retrieval (finding the relevant articles) with that of machine comprehension of text (identifying the answer spans from those articles). Our approach combines a search component based on bigram hashing and TF-IDF matching with a multi-layer recurrent neural network model trained to detect answers in Wikipedia paragraphs. Read More

2017Mar
Authors: MicroBooNE collaboration, P. Abratenko, R. Acciarri, C. Adams, R. An, J. Asaadi, M. Auger, L. Bagby, S. Balasubramanian, B. Baller, C. Barnes, G. Barr, M. Bass, F. Bay, M. Bishai, A. Blake, T. Bolton, L. Bugel, L. Camilleri, D. Caratelli, B. Carls, R. Castillo Fernandez, F. Cavanna, H. Chen, E. Church, D. Cianci, E. Cohen, G. H. Collin, J. M. Conrad, M. Convery, J. I. Crespo-Anadon, M. Del Tutto, D. Devitt, S. Dytman, B. Eberly, A. Ereditato, L. Escudero Sanchez, J. Esquivel, B. T. Fleming, W. Foreman, A. P. Furmanski, D. Garcia-Gamez, G. T. Garvey, V. Genty, D. Goeldi, S. Gollapinni, N. Graf, E. Gramellini, H. Greenlee, R. Grosso, R. Guenette, A. Hackenburg, P. Hamilton, O. Hen, J. Hewes, C. Hill, J. Ho, G. Horton-Smith, E. -C. Huang, C. James, J. Jan de Vries, C. -M. Jen, L. Jiang, R. A. Johnson, B. J. P. Jones, J. Joshi, H. Jostlein, D. Kaleko, L. N. Kalousis, G. Karagiorgi, W. Ketchum, B. Kirby, M. Kirby, T. Kobilarcik, I. Kreslo, A. Laube, Y. Li, A. Lister, B. R. Littlejohn, S. Lockwitz, D. Lorca, W. C. Louis, M. Luethi, B. Lundberg, X. Luo, A. Marchionni, C. Mariani, J. Marshall, D. A. Martinez Caicedo, V. Meddage, T. Miceli, G. B. Mills, J. Moon, M. Mooney, C. D. Moore, J. Mousseau, R. Murrells, D. Naples, P. Nienaber, J. Nowak, O. Palamara, V. Paolone, V. Papavassiliou, S. F. Pate, Z. Pavlovic, E. Piasetzky, D. Porzio, G. Pulliam, X. Qian, J. L. Raaf, A. Rafique, L. Rochester, C. Rudolf von Rohr, B. Russell, D. W. Schmitz, A. Schukraft, W. Seligman, M. H. Shaevitz, J. Sinclair, E. L. Snider, M. Soderberg, S. Soldner-Rembold, S. R. Soleti, P. Spentzouris, J. Spitz, J. St. John, T. Strauss, A. M. Szelc, N. Tagg, K. Terao, M. Thomson, M. Toups, Y. -T. Tsai, S. Tufanli, T. Usher, R. G. Van de Water, B. Viren, M. Weber, J. Weston, D. A. Wickremasinghe, S. Wolbers, T. Wongjirad, K. Woodruff, T. Yang, L. Yates, G. P. Zeller, J. Zennamo, C. Zhang

We discuss a technique for measuring a charged particle's momentum by means of multiple Coulomb scattering (MCS) in the MicroBooNE liquid argon time projection chamber (LArTPC). This method does not require the full particle ionization track to be contained inside of the detector volume as other track momentum reconstruction methods do (range-based momentum reconstruction and calorimetric momentum reconstruction). We motivate use of this technique, describe a tuning of the underlying phenomenological formula, quantify its performance on fully contained beam-neutrino-induced muon tracks both in simulation and in data, and quantify its performance on exiting muon tracks in simulation. Read More

2017Mar
Affiliations: 1The George Washington University, 2Michigan State University, 3University of Pittsburgh, 4Michigan State University, 5Stony Brook University, 6Columbia University, 7NASA/Goddard, 8NRAO, 9The George Washington University, 10GBO, 11Herzberg Institute of Astrophysics

We present multi-wavelength observations of the unusual nova V1535 Sco throughout its outburst in 2015. Early radio observations were consistent with synchrotron emission, and early X-ray observations revealed the presence of high-energy (>1 keV) photons. These indicated that strong shocks were present during the first ~2 weeks of the nova's evolution. Read More

It has recently been discovered that some, if not all, classical novae emit GeV gamma-rays during outburst, but the mechanics of this gamma-ray emission are still not well understood. We present here a comprehensive, multi-wavelength dataset---from radio to X-rays---for the most gamma-ray luminous classical nova to-date, V1324 Sco. Using this dataset, we show that V1324 Sco is a canonical dusty Fe-II type nova, with a bulk ejecta velocity of $1150 \pm 40~\rm km~s^{-1}$ and an ejecta mass of $2. Read More

2016Dec
Authors: MicroBooNE Collaboration, R. Acciarri, C. Adams, R. An, A. Aparicio, S. Aponte, J. Asaadi, M. Auger, N. Ayoub, L. Bagby, B. Baller, R. Barger, G. Barr, M. Bass, F. Bay, K. Biery, M. Bishai, A. Blake, V. Bocean, D. Boehnlein, V. D. Bogert, T. Bolton, L. Bugel, C. Callahan, L. Camilleri, D. Caratelli, B. Carls, R. Castillo Fernandez, F. Cavanna, S. Chappa, H. Chen, K. Chen, C. Y. Chi, C. S. Chiu, E. Church, D. Cianci, G. H. Collin, J. M. Conrad, M. Convery, J. Cornele, P. Cowan, J. I. Crespo-Anadon, G. Crutcher, C. Darve, R. Davis, M. Del Tutto, D. Devitt, S. Duffin, S. Dytman, B. Eberly, A. Ereditato, D. Erickson, L. Escudero Sanchez, J. Esquivel, S. Farooq, J. Farrell, D. Featherston, B. T. Fleming, W. Foreman, A. P. Furmanski, V. Genty, M. Geynisman, D. Goeldi, B. Goff, S. Gollapinni, N. Graf, E. Gramellini, J. Green, A. Greene, H. Greenlee, T. Griffin, R. Grosso, R. Guenette, A. Hackenburg, R. Haenni, P. Hamilton, P. Healey, O. Hen, E. Henderson, J. Hewes, C. Hill, K. Hill, L. Himes, J. Ho, G. Horton-Smith, D. Huffman, C. M. Ignarra, C. James, E. James, J. Jan de Vries, W. Jaskierny, C. M. Jen, L. Jiang, B. Johnson, M. Johnson, R. A. Johnson, B. J. P. Jones, J. Joshi, H. Jostlein, D. Kaleko, L. N. Kalousis, G. Karagiorgi, T. Katori, P. Kellogg, W. Ketchum, J. Kilmer, B. King, B. Kirby, M. Kirby, E. Klein, T. Kobilarcik, I. Kreslo, R. Krull, R. Kubinski, G. Lange, F. Lanni, A. Lathrop, A. Laube, W. M. Lee, Y. Li, D. Lissauer, A. Lister, B. R. Littlejohn, S. Lockwitz, D. Lorca, W. C. Louis, G. Lukhanin, M. Luethi, B. Lundberg, X. Luo, G. Mahler, I. Majoros, D. Makowiecki, A. Marchionni, C. Mariani, D. Markley, J. Marshall, D. A. Martinez Caicedo, K. T. McDonald, D. McKee, A. McLean, J. Mead, V. Meddage, T. Miceli, G. B. Mills, W. Miner, J. Moon, M. Mooney, C. D. Moore, Z. Moss, J. Mousseau, R. Murrells, D. Naples, P. Nienaber, B. Norris, N. Norton, J. Nowak, M. OBoyle, T. Olszanowski, O. Palamara, V. Paolone, V. Papavassiliou, S. F. Pate, Z. Pavlovic, R. Pelkey, M. Phipps, S. Pordes, D. Porzio, G. Pulliam, X. Qian, J. L. Raaf, V. Radeka, A. Rafique, R. A Rameika, B. Rebel, R. Rechenmacher, S. Rescia, L. Rochester, C. Rudolf von Rohr, A. Ruga, B. Russell, R. Sanders, W. R. Sands III, M. Sarychev, D. W. Schmitz, A. Schukraft, R. Scott, W. Seligman, M. H. Shaevitz, M. Shoun, J. Sinclair, W. Sippach, T. Smidt, A. Smith, E. L. Snider, M. Soderberg, M. Solano-Gonzalez, S. Soldner-Rembold, S. R. Soleti, J. Sondericker, P. Spentzouris, J. Spitz, J. St. John, T. Strauss, K. Sutton, A. M. Szelc, K. Taheri, N. Tagg, K. Tatum, J. Teng, K. Terao, M. Thomson, C. Thorn, J. Tillman, M. Toups, Y. T. Tsai, S. Tufanli, T. Usher, M. Utes, R. G. Van de Water, C. Vendetta, S. Vergani, E. Voirin, J. Voirin, B. Viren, P. Watkins, M. Weber, T. Wester, J. Weston, D. A. Wickremasinghe, S. Wolbers, T. Wongjirad, K. Woodruff, K. C. Wu, T. Yang, B. Yu, G. P. Zeller, J. Zennamo, C. Zhang, M. Zuckerbrot

This paper describes the design and construction of the MicroBooNE liquid argon time projection chamber and associated systems. MicroBooNE is the first phase of the Short Baseline Neutrino program, located at Fermilab, and will utilize the capabilities of liquid argon detectors to examine a rich assortment of physics topics. In this document details of design specifications, assembly procedures, and acceptance tests are reported. Read More

A good dialogue agent should have the ability to interact with users by both responding to questions and by asking questions, and importantly to learn from both types of interaction. In this work, we explore this direction by designing a simulator and a set of synthetic tasks in the movie domain that allow such interactions between a learner and a teacher. We investigate how a learner can benefit from asking questions in both offline and online reinforcement learning settings, and demonstrate that the learner improves when asking questions. Read More

We introduce a new model, the Recurrent Entity Network (EntNet). It is equipped with a dynamic long-term memory which allows it to maintain and update a representation of the state of the world as it receives new data. For language understanding tasks, it can reason on-the-fly as it reads text, not just when it is required to answer a question or respond as is the case for a Memory Network (Sukhbaatar et al. Read More

An important aspect of developing conversational agents is to give a bot the ability to improve through communicating with humans and to learn from the mistakes that it makes. Most research has focused on learning from fixed training sets of labeled data rather than interacting with a dialogue partner in an online fashion. In this paper we explore this direction in a reinforcement learning setting where the bot improves its question-answering ability from feedback a teacher gives following its generated responses. Read More

2016Nov
Authors: MicroBooNE collaboration, R. Acciarri, C. Adams, R. An, J. Asaadi, M. Auger, L. Bagby, B. Baller, G. Barr, M. Bass, F. Bay, M. Bishai, A. Blake, T. Bolton, L. Bugel, L. Camilleri, D. Caratelli, B. Carls, R. Castillo Fernandez, F. Cavanna, H. Chen, E. Church, D. Cianci, G. H. Collin, J. M. Conrad, M. Convery, J. I. Crespo-Anadón, M. Del Tutto, D. Devitt, S. Dytman, B. Eberly, A. Ereditato, L. Escudero Sanchez, J. Esquivel, B. T. Fleming, W. Foreman, A. P. Furmanski, G. T. Garvey, V. Genty, D. Goeldi, S. Gollapinni, N. Graf, E. Gramellini, H. Greenlee, R. Grosso, R. Guenette, A. Hackenburg, P. Hamilton, O. Hen, J. Hewes, C. Hill, J. Ho, G. Horton-Smith, C. James, J. Jan de Vries, C. -M. Jen, L. Jiang, R. A. Johnson, B. J. P. Jones, J. Joshi, H. Jostlein, D. Kaleko, G. Karagiorgi, W. Ketchum, B. Kirby, M. Kirby, T. Kobilarcik, I. Kreslo, A. Laube, Y. Li, A. Lister, B. R. Littlejohn, S. Lockwitz, D. Lorca, W. C. Louis, M. Luethi, B. Lundberg, X. Luo, A. Marchionni, C. Mariani, J. Marshall, D. A. Martinez Caicedo, V. Meddage, T. Miceli, G. B. Mills, J. Moon, M. Mooney, C. D. Moore, J. Mousseau, R. Murrells, D. Naples, P. Nienaber, J. Nowak, O. Palamara, V. Paolone, V. Papavassiliou, S. F. Pate, Z. Pavlovic, D. Porzio, G. Pulliam, X. Qian, J. L. Raaf, A. Rafique, L. Rochester, C. Rudolf von Rohr, B. Russell, D. W. Schmitz, A. Schukraft, W. Seligman, M. H. Shaevitz, J. Sinclair, E. L. Snider, M. Soderberg, S. Söldner-Rembold, S. R. Soleti, P. Spentzouris, J. Spitz, J. St. John, T. Strauss, A. M. Szelc, N. Tagg, K. Terao, M. Thomson, M. Toups, Y. -T. Tsai, S. Tufanli, T. Usher, R. G. Van de Water, B. Viren, M. Weber, J. Weston, D. A. Wickremasinghe, S. Wolbers, T. Wongjirad, K. Woodruff, T. Yang, G. P. Zeller, J. Zennamo, C. Zhang

We present several studies of convolutional neural networks applied to data coming from the MicroBooNE detector, a liquid argon time projection chamber (LArTPC). The algorithms studied include the classification of single particle images, the localization of single particle and neutrino interactions in an image, and the detection of a simulated neutrino event overlaid with cosmic ray backgrounds taken from real detector data. These studies demonstrate the potential of convolutional neural networks for particle identification or event detection on simulated neutrino interactions. Read More

Directly reading documents and being able to answer questions from them is an unsolved challenge. To avoid its inherent difficulty, question answering (QA) has been directed towards using Knowledge Bases (KBs) instead, which has proven effective. Unfortunately KBs often suffer from being too restrictive, as the schema cannot support certain types of answers, and too sparse, e. Read More

Traditional dialog systems used in goal-oriented applications require a lot of domain-specific handcrafting, which hinders scaling up to new domains. End-to-end dialog systems, in which all components are trained from the dialogs themselves, escape this limitation. But the encouraging success recently obtained in chit-chat dialog may not carry over to goal-oriented settings. Read More

A long-term goal of machine learning research is to build an intelligent dialog agent. Most research in natural language understanding has focused on learning from fixed training sets of labeled data, with supervision either at the word level (tagging, parsing tasks) or sentence level (question answering, machine translation). This kind of supervision is not realistic of how humans learn, where language is both learned by, and used for, communication. Read More

We report on our recent efforts to perform realistic simulations of large quantum devices in the time domain. In contrast to d.c. Read More

A long-term goal of machine learning is to build intelligent conversational agents. One recent popular approach is to train end-to-end models on a large amount of real dialog transcripts between humans (Sordoni et al., 2015; Vinyals & Le, 2015; Shang et al. Read More

We introduce a new test of how well language models capture meaning in children's books. Unlike standard language modelling benchmarks, it distinguishes the task of predicting syntactic function words from that of predicting lower-frequency words, which carry greater semantic content. We compare a range of state-of-the-art models, each with a different way of encoding what has been previously read. Read More

Since the Fermi discovery of $\gamma$-rays from novae, one of the biggest questions in the field has been how novae generate such high-energy emission. Shocks must be a fundamental ingredient. Six months of radio observations of the 2012 nova V5589 Sgr with the VLA and 15 weeks of X-ray observations with Swift/XRT show that the radio emission consisted of: 1) a shock-powered, non-thermal flare; and 2) weak thermal emission from $10^{-5}$ M$_\odot$ of freely expanding, photoionized ejecta. Read More

We report on a "source-sink" algorithm which allows one to calculate time-resolved physical quantities from a general nanoelectronic quantum system (described by an arbitrary time-dependent quadratic Hamiltonian) connected to infinite electrodes. Although mathematically equivalent to the non equilibrium Green's function formalism, the approach is based on the scattering wave functions of the system. It amounts to solving a set of generalized Schr\"odinger equations which include an additional "source" term (coming from the time dependent perturbation) and an absorbing "sink" term (the electrodes). Read More

Harris Corporation has an interest in making HF radios a suitable medium for wireless information networks using standard Internet protocols. Although HF radio links have many unique characteristics, HF wireless subnets can be subject to many of the same traffic flow characteristics and topologies as existing line-of-sight (LOS) radio networks, giving rise to similar issues (media access, connectivity, routing) which lend themselves to investigation through simulation. Accordingly, we have undertaken to develop efficient, high-fidelity simulations of various aspects of HF radio communications and networking using the OMNeT++ framework. Read More

Summarization based on text extraction is inherently limited, but generation-style abstractive methods have proven challenging to build. In this work, we propose a fully data-driven approach to abstractive sentence summarization. Our method utilizes a local attention-based model that generates each word of the summary conditioned on the input sentence. Read More

We consider the differential conductance of a periodically driven system connected to infinite electrodes. We focus on the situation where the dissipation occurs predominantly in these electrodes. Using analytical arguments and a detailed numerical study we relate the differential conductances of such a system in two and three terminal geometries to the spectrum of quasi-energies of the Floquet operator. Read More

Training large-scale question answering systems is complicated because training sources usually cover a small portion of the range of possible questions. This paper studies the impact of multitask and transfer learning for simple question answering; a setting for which the reasoning required to answer is quite easy, as long as one can retrieve the correct evidence given a question, which can be difficult in large-scale conditions. To this end, we introduce a new dataset of 100k questions that we use in conjunction with existing benchmarks. Read More

The importance of shocks in nova explosions has been highlighted by Fermi's discovery of \gamma-ray producing novae. Over three years of multi-band VLA radio observations of the 2010 nova V1723 Aql show that shocks between fast and slow flows within the ejecta led to the acceleration of particles and the production of synchrotron radiation. Soon after the start of the eruption, shocks in the ejecta produced an unexpected radio flare, resulting in a multi-peaked radio light curve. Read More

We introduce a neural network with a recurrent attention model over a possibly large external memory. The architecture is a form of Memory Network (Weston et al., 2015) but unlike the model in that work, it is trained end-to-end, and hence requires significantly less supervision during training, making it more generally applicable in realistic settings. Read More

Determining reliable distances to classical novae is a challenging but crucial step in deriving their ejected masses and explosion energetics. Here we combine radio expansion measurements from the Karl G. Jansky Very Large Array with velocities derived from optical spectra to estimate an expansion parallax for nova V959 Mon, the first nova discovered through its gamma-ray emission. Read More

One long-term goal of machine learning research is to produce methods that are applicable to reasoning and natural language, in particular building an intelligent dialogue agent. To measure progress towards that goal, we argue for the usefulness of a set of proxy tasks that evaluate reading comprehension via question answering. Our tasks measure understanding in several ways: whether a system is able to answer questions via chaining facts, simple induction, deduction and many more. Read More

We study the interplay between Andreev (Majorana) bound states that form at the boundary of a (topological) superconductor and a train of microwave pulses. We find that the extra dynamical phase coming from the pulses can shift the phase of the Andreev reflection, resulting in the appear- ance of dynamical Andreev states. As an application we study the presence of the zero bias peak in the differential conductance of a normal-topological superconductor junction - the simplest, yet somehow ambiguous, experimental signature for Majorana states. Read More

We describe a new class of learning models called memory networks. Memory networks reason with inference components combined with a long-term memory component; they learn how to use these jointly. The long-term memory can be read and written to, with the goal of using it for prediction. Read More

Classical novae are the most common astrophysical thermonuclear explosions, occurring on the surfaces of white dwarf stars accreting gas from companions in binary star systems. Novae typically expel ~10^(-4) solar masses of material at velocities exceeding 1,000 kilometres per second. However, the mechanism of mass ejection in novae is poorly understood, and could be dominated by the impulsive flash of thermonuclear energy, prolonged optically thick winds, or binary interaction with the nova envelope. Read More

Superconductivity derives its most salient features from the coherence of its macroscopic wave function. The associated physical phenomena have now moved from exotic subjects to fundamental building blocks for quantum circuits such as qubits or single photonic modes. Here, we theoretically find that the AC Josephson effect---which transforms a DC voltage $V_b$ into an oscillating signal $cos(2eV_b t/ \hbar)$---has a mesoscopic counterpart in normal conductors. Read More

With the technical progress of radio-frequency setups, high frequency quantum transport experiments have moved from theory to the lab. So far the standard theoretical approach used to treat such problems numerically--known as Keldysh or NEGF (Non Equilibrium Green's Functions) formalism--has not been very successful mainly because of a prohibitive computational cost. We propose a reformulation of the non-equilibrium Green's function technique in terms of the electronic wave functions of the system in an energy-time representation. Read More

This paper presents a system which learns to answer questions on a broad range of topics from a knowledge base using few hand-crafted features. Our model learns low-dimensional embeddings of words and knowledge base constituents; these representations are used to score natural language questions against candidate answers. Training our system using pairs of questions and structured representations of their answers, and pairs of question paraphrases, yields competitive results on a competitive benchmark of the literature. Read More

Most functionalities of modern electronic circuits rely on the possibility to modify the path fol- lowed by the electrons using, e.g. field effect transistors. Read More

Building computers able to answer questions on any subject is a long standing goal of artificial intelligence. Promising progress has recently been achieved by methods that learn to map questions to logical forms or database queries. Such approaches can be effective but at the cost of either large amounts of human-labeled data or by defining lexicons and grammars tailored by practitioners. Read More

The recurrent nova T Pyx underwent its sixth historical outburst in 2011, and became the subject of an intensive multi-wavelength observational campaign. We analyze data from the Swift and Suzaku satellites to produce a detailed X-ray light curve augmented by epochs of spectral information. X-ray observations yield mostly non-detections in the first four months of outburst, but both a super-soft and hard X-ray component rise rapidly after Day 115. Read More

This paper proposes a novel approach for relation extraction from free text which is trained to jointly use information from the text and from existing knowledge. Our model is based on two scoring functions that operate by learning low-dimensional embeddings of words and of entities and relationships from a knowledge base. We empirically show on New York Times articles aligned with Freebase relations that our approach is able to efficiently use the extra information provided by a large subset of Freebase data (4M entities, 23k relationships) to improve over existing methods that rely on text features alone. Read More

This paper discusses the technical aspects - mathematical and numerical - associated with the numerical simulations of a mesoscopic system in the time domain (i.e. beyond the single frequency AC limit). Read More

The radio light curves of novae rise and fall over the course of months to years, allowing for detailed observations of the evolution of the nova shell. However, the main parameter determined by radio models of nova explosions - the mass of the ejecta - often seems to exceed theoretical expectations by an order of magnitude. With the recent technological improvements on the Karl G. Read More

We consider the problem of embedding entities and relations of knowledge bases in low-dimensional vector spaces. Unlike most existing approaches, which are primarily efficient for modeling equivalence relations, our approach is designed to explicitly model irreflexive relations, such as hierarchies, by interpreting them as translations operating on the low-dimensional embeddings of the entities. Preliminary experiments show that, despite its simplicity and a smaller number of parameters than previous approaches, our approach achieves state-of-the-art performance according to standard evaluation protocols on data from WordNet and Freebase. Read More

Novae, which are the sudden visual brightening triggered by runaway thermonuclear burning on the surface of an accreting white dwarf, are fairly common and bright events. Despite their astronomical significance as nearby laboratories for the study of nuclear burning and accretion phenomena, many aspects of these common stellar explosions are observationally not well-constrained and remain poorly understood. Radio observations, modeling and interpretation can potentially play a crucial role in addressing some of these puzzling issues. Read More

Supervised (linear) embedding models like Wsabie and PSI have proven successful at ranking, recommendation and annotation tasks. However, despite being scalable to large datasets they do not take full advantage of the extra data due to their linear nature, and typically underfit. We propose a new class of models which aim to provide improved performance while retaining many of the benefits of the existing class of embedding models. Read More

Large-scale relational learning becomes crucial for handling the huge amounts of structured data generated daily in many application domains ranging from computational biology or information retrieval, to natural language processing. In this paper, we present a new neural network architecture designed to embed multi-relational graphs into a flexible continuous vector space in which the original data is kept and enhanced. The network is trained to encode the semantics of these graphs in order to assign high probabilities to plausible components. Read More

Despite being the prototype of its class, T Pyx is arguably the most unusual and poorly understood recurrent nova. Here, we use radio observations from the Karl G. Jansky Very Large Array to trace the evolution of the ejecta over the course of the 2011 outburst of T Pyx. Read More

We present multi-frequency radio observations of the 2010 nova event in the symbiotic binary V407 Cygni, obtained with the Karl G. Jansky Very Large Array and spanning 1-45 GHz and 17-770 days following discovery. This nova---the first ever detected in gamma rays---shows a radio light curve dominated by the wind of the Mira giant companion, rather than the nova ejecta themselves. Read More

Many latent (factorized) models have been proposed for recommendation tasks like collaborative filtering and for ranking tasks like document or image retrieval and annotation. Common to all those methods is that during inference the items are scored independently by their similarity to the query in the latent embedding space. The structure of the ranked list (i. Read More

Retrieval tasks typically require a ranking of items given a query. Collaborative filtering tasks, on the other hand, learn to model user's preferences over items. In this paper we study the joint problem of recommending items to a user with respect to a given query, which is a surprisingly common task. Read More

Open-text (or open-domain) semantic parsers are designed to interpret any statement in natural language by inferring a corresponding meaning representation (MR). Unfortunately, large scale systems cannot be easily machine-learned due to lack of directly supervised data. We propose here a method that learns to assign MRs to a wide range of text (using a dictionary of more than 70,000 words, which are mapped to more than 40,000 entities) thanks to a training scheme that combines learning from WordNet and ConceptNet with learning from raw text. Read More

Music prediction tasks range from predicting tags given a song or clip of audio, predicting the name of the artist, or predicting related songs given a song, clip, artist name or tag. That is, we are interested in every semantic relationship between the different musical concepts in our database. In realistically sized databases, the number of songs is measured in the hundreds of thousands or more, and the number of artists in the tens of thousands or more, providing a considerable challenge to standard machine learning techniques. Read More

We propose a unified neural network architecture and learning algorithm that can be applied to various natural language processing tasks including: part-of-speech tagging, chunking, named entity recognition, and semantic role labeling. This versatility is achieved by trying to avoid task-specific engineering and therefore disregarding a lot of prior knowledge. Instead of exploiting man-made input features carefully optimized for each task, our system learns internal representations on the basis of vast amounts of mostly unlabeled training data. Read More