Abstract

There have been numerous attempts in explaining the general learning behaviours using model-based and model-free methods. While the model-based control is flexible yet computationally expensive in planning, the model-free control is quick but inflexible. Multiple arbitration schemes have been suggested to achieve the data efficiency and computational efficiency of model-based and model-free control schemes, respectively. In this context, we propose a quantitative ’valueof-information’ based arbitration between both the controllers in order to establish a general computational framework for skill learning. The interacting model-based and model-free reinforcement learning processes are arbitrated using an uncertainty-based value-of-information estimation. We further show that our algorithm performs better than Q-learning as well as Q-learning with experience replay.

Abstract

Several canonical experimental paradigms (serial reaction task, mxn task, etc.) have been proposed to study the typical behavioural phenomenon in sequential key-press tasks. However, not much work has been done on studying motor sequencing in grid-navigation tasks. In this work, using empirical examinations, we systematically show grid-navigation task as an instance of skill learning paradigm. The participants performed Grid-Sailing Task (GST), which required navigating (by executing sequential key-presses) a 5x5 grid from start to goal position while using a particular key-mapping among the 3 cursor movement directions and the 3 keyboard buttons. We employ two different experiments to argue for the learning of cognitive strategies as well as motor sequences. By rejecting the motor adaptation argument and validating the law of practice, we characterize GST as a skill learning task. We further argue for advantages of GST as a general, canonical task over others for use in skill learning studies.

Abstract

Several canonical experimental paradigms (serial reaction task, mxn task, etc.) have been proposed to study the typical behavioural phenomena in a sequential motor key-press task. The repeated execution of visuomotor sequences in such paradigms lead to overall performance improvement such that the inter-response intervals in between certain sub-sequences decreases as compared to that across other sub-sequences. This efficient and hierarchical cluster organisation is called \textit{motor chunking}. We provide empirical evidence for motor chunking in grid-navigation sequencing tasks. The participants performed Grid-Sailing Task (GST) [Fermin et. al., 2010] that required navigating a 10x10 grid from start to goal position while using a particular key-mapping between the 3 cursor movement directions and the 3 keyboard buttons. This study confirms the emergence of subject-specific, unique temporal patterns related to chunking after substantial practice.

Abstract

—Social media platforms have taken political and cultural conversations to an online platform making them more accessible. Ability to anonymously post has allowed more people to participate fearlessly. However, this has also led to an opportunity to spread miss information and manipulative content. Political groups around the globe have used Bot accounts to help spread their preferred narrative online during elections. In the midst of 2019 Indian Lok Sabha Elections speculations were made about the presence of cyber-troops/IT Cells which operate fake accounts and push propaganda. Our finding suggests that a portion of Bot accounts seems to be operated by humans in the background. These accounts have a very distinct usage pattern on Twitter compared to legitimate human users. Our experiments also point out that only 1.3% of total interactions are directed from Humans to Bots, showing Bot accounts inability to gel well in the online social network. Keywords-Social Networks, Bots, Twitter, Elections

Abstract

Given that India is now moving towards automated solutions to curb traffic violations and road accidents, we focus our efforts on characterizing these violations in Indian cities. In this work, we present our characterization of the traffic violations via an Automated e-challan (electronic trafficviolation receipt) issuance system of Ahmedabad and New Delhi. To explore this, we collected an exhaustive dataset of over 6 million e-challans. Characterizing the fine payment behavior, we find that 57% of unique vehicles in Ahmedabad are involved in repeat offenses. The temporal analysis shows a significant difference in e-challans issued during the festivals. Spatially, different violation types are distributed differently with the existence of certain unique hotspots. Finally, we also demonstrate how e-challans can act as a proxy measure to analyze the efficacy of the Motor Vehicles (Amendment) Act 2019. Our work suggests that high penalties may not have a long term impact on decreasing traffic violations. Keywords-big data; traffic violations; e-challan; transport systems; state policy; surveillance;

Abstract

In 2015, 391,000 people were injured due to distracted driv-ing in the US. One of the major reasons behind distracteddriving is the use of cell-phones, accounting for 14% of fatalcrashes. Social media applications have enabled users to stayconnected, however, the use of such applications while driv-ing could have serious repercussions - often leading the userto be distracted from the road and ending up in an accident.In the context of impression management, it has been discov-ered that individuals often take a risk (such as teens smokingcigarettes, indulging in narcotics, and participating in unsafesex) to improve their social standing. Therefore, viewing thephenomena of posting distracted driving posts under the lensof self-presentation, it can be hypothesized that users oftenindulge in risk-taking behavior on social media to improvetheir impression among their peers. In this paper, we firsttry to understand the severity of such social-media-based dis-tractions by analyzing the content posted on a popular socialmedia site where the user is driving and is also simultane-ously creating content. To this end, we build a deep learningclassifier to identify publicly posted content on social mediathat involves the user driving. Furthermore, a framework pro-posed to understand factors behind voluntary risk-taking ac-tivity observes that younger individuals are more willing toperform such activities, and men (as opposed to women) aremore inclined to take risks. Grounding our observations inthis framework, we test these hypotheses on 173 cities acrossthe world. We conduct spatial and temporal analysis on a city-level and understand how distracted driving content postingbehavior changes due to varied demographics. We discoverthat the factors put forth by the framework are significant inestimating the extent of such behavior.

Abstract

With the rise of social networking platforms, everyone now has free access to information from around the work. Anyone from anywhere can now share context with the entire world. This allows for more connectivity around the world and more transparency. However, this also allows for the spread of misinformation and fake news often resulting in undesired and extremely impactful political, economic, social, psychological and criminal consequences. Identifying the fake news spreaders is as important as identifying the fake news itself. We put forward a method to utilize content analysis and more user modelling to capture who is more likely to share fake news. We use TF-IDF as our text transformation method coupled with algorithms simple classification algorithm Logistic Regression and achieve an accuracy of 71.5% and 70% in identifying fake news spreaders in both the English as well as Spanish test set respectively.

Abstract

Discourse relations between two textual spans in a document attempt to capture the coherent structure which emerges in language use. Automatic classification of these relations remains a challenging task especially in case of implicit discourse relations, where there is no explicit textual cue which marks the discourse relation. In low resource languages, this motivates the exploration of transfer learning approaches, more particularly the cross-lingual techniques towards discourse relation classification. In this work, we explore various cross-lingual transfer techniques on Hindi Discourse Relation Bank (HDRB), a Penn Discourse Treebank styled dataset for discourse analysis in Hindi and observe performance gains in both zero shot and finetuning settings on the Hindi Discourse Relation Classification task. This is the first effort towards exploring transfer learning for Hindi Discourse relation classification to the best of our knowledge.

Abstract

High-temperature weather conditions induce geometrical distortions in images which can adversely affect the performance of a computer vision model performing downstream tasks such as semantic segmentation. The performance of such models has been shown to improve by adding a restoration network before a semantic segmentation network. The restoration network removes the geometrical distortions from the images and shows improved segmentation results. However, this approach suffers from a major architec- tural drawback that is the restoration network does not learn directly from the errors of the segmentation network. In other words, the restoration network is not task aware. In this work, we propose a semantic feedback learning approach, which improves the task of semantic segmentation giving a feedback response into the restoration network. This response works as an attend and fix mechanism by focusing on those areas of an image where restoration needs improvement. Also, we proposed loss functions: Iterative Fo- cal Loss (iFL) and Class-Balanced Iterative Focal Loss (CB-iFL), which are specifically designed to improve the performance of the feedback network. These losses focus more on those samples that are continuously miss-classified over successive iterations. Our approach gives a gain of 17.41 mIoU over the standard segmentation model, including the additional gain of 1.9 mIoU with CB-iFL on the Cityscapes dataset.

Abstract

Inspired by the success of the CNN-RNN framework in the image captioning task, several works have explored this in multi-label image annotation with the hope that the RNN followed by a CNN would encode inter-label dependencies better than using a CNN alone. To do so, for each training sample, the earlier methods converted the ground- truth label-set into a sequence of labels based on their frequencies (e.g., rare-to-frequent) for training the RNN. However, since the ground-truth is an unordered set of labels, imposing a fixed and predefined sequence on them does not naturally align with this task. To address this, some of the recent papers have proposed techniques that are capable to train the RNN without feeding the ground-truth labels in a particular se- quence/order. However, most of these techniques leave it to the RNN to implicitly choose one sequence for the ground-truth labels corresponding to each sample at the time of training, thus making it inherently biased. In this paper, we address this limitation and propose a novel approach in which the RNN is explicitly forced to learn multiple relevant inter- label dependencies, without the need of feeding the ground-truth in any particular order. Using thorough empirical comparisons, we demonstrate that our approach outperforms several state-of-the-art techniques on two popular datasets (MS-COCO and NUS-WIDE). Additionally, it provides a new perspecitve of looking at an unordered set of labels as equivalent to a collection of different permutations (sequences) of those labels, thus naturally aligning with the image annotation task. Our code is available