Sandeep Subramanian

Neural architectures for named entity recognition

Comunicacio presentada a la 2016 Conference of the North American Chapter of the Association for Computational Linguistics, celebrada a San Diego (CA, EUA) els dies 12 a 17 de juny 2016.

Adversarial Generation of Natural Language.

Generative Adversarial Networks (GANs) have gathered a lot of attention from the computer vision community, yielding impressive results for image generation. Advances in the adversarial generation of natural language from noise however are not commensurate with the progress made in generating images, and still lag far behind likelihood based methods. In this paper, we take a step towards generating natural language with a GAN objective alone. We introduce a simple baseline that addresses the discrete output space problem without relying on gradient estimators and show that it is able to achieve state-of-the-art results on a Chinese poem generation dataset. We present quantitative results on generating sentences from context-free and probabilistic context-free grammars, and qualitative language modeling results. A conditional version is also described that can generate sequences conditioned on sentence characteristics.

Machine Comprehension by Text-to-Text Neural Question Generation

We propose a recurrent neural model that generates natural-language questions from documents, conditioned on answers. We show how to train the model using a combination of supervised and reinforcement learning. After teacher forcing for standard maximum likelihood training, we fine-tune the model using policy gradient techniques to maximize several rewards that measure question quality. Most notably, one of these rewards is the performance of a question-answering system. We motivate question generation as a means to improve the performance of question answering systems. Our model is trained and evaluated on the recent question-answering dataset SQuAD.

Influence of training environment and cognitive deficits on use of feedback for motor learning in chronic stroke

Feedback provision is an essential component of motor learning for improving upper limb recovery in people with stroke. Along with sensorimotor impairments, many individuals post-stroke have cognitive deficits that can influence arm motor recovery. However, few studies have identified whether the training environment and presence of cognitive deficits influences the ability to use feedback in individuals post-stroke. We evaluated the influence of the training environment and cognitive impairments on the ability to use feedback to enhance arm motor recovery. Twenty-four subjects with chronic post-stroke upper limb hemiparesis were randomized to practice pointing movements in a 3D virtual environment (VE) or a similarly designed physical environment (PE; n=12/group) for 12 sessions (72 trials/session, 3 days/week). All participants were provided with feedback about movement speed (Knowledge of Results) and trunk displacement (Knowledge of Performance). Neurocognitive functioning was assessed only before task practice (PRE), while kinematic assessments were carried out at PRE, immediately after (POST) and 3 months (RET) after task practice. Repeated measures ANOVAs with mixed models assessed the changes in kinematic outcomes. Neurocognitive function was correlated with kinematic outcomes. Those training in the VE had greater endpoint speed and ranges of shoulder horizontal adduction, shoulder flexion and elbow extension. They also tended to use less trunk displacement. Kinematic deficits influenced the ability to use feedback in people with chronic stroke for upper limb motor learning and recovery. Information about the presence of these deficits can help in the selection of the most appropriate interventions for maximizing arm motor recovery and motor learning in chronic stroke.

Incorporation of motor control and motor learning principles into VR applications

The primary focus of rehabilitation for individuals with motor deficits is the relearning of specific motor skills and daily tasks. Rehabilitation strives to take advantage of neuroplastic processes during recovery, a process that can be addressed by creating enriched training environments using virtual reality (VR) based simulations. The objectives of this workshop are to review motor control and motor learning principles, to discuss how they can be exploited by VR training environments and to provide examples of how these principles have been incorporated into different VR simulations for improving upper limb motor recovery. The workshop includes a practical component in which participants will design a specific intervention for improving a typical motor problem incorporating motor control and motor learning principles, in both a VR-based and a non-VR-based clinical application. Finally, we will discuss the limitations of the current technologies with respect to their effectiveness and transfer of learning to daily life tasks.