In this paper we investigate two hypothesis regarding the use of deep reinforcement learning in multiple tasks. The first hypothesis is driven by the question of whether a deep reinforcement learning algorithm, trained on two similar tasks, is able to outperform two single-task, individually trained algorithms, by more efficiently learning a new, similar task, that none of the three algorithms has encountered before. The second hypothesis is driven by the question of whether the same multi-task deep RL algorithm, trained on two similar tasks and augmented with elastic weight consolidation (EWC), is able to retain similar performance on the new task, as a similar algorithm without EWC, whilst being able to overcome catastrophic forgetting in the two previous tasks. We show that a multi-task Asynchronous Advantage Actor-Critic (GA3C) algorithm, trained on Space Invaders and Demon Attack, is in fact able to outperform two single-tasks GA3C versions, trained individually for each single-task, when evaluated on a new, third task, namely, Phoenix. We also show that, when training two trained multi-task GA3C algorithms on the third task, if one is augmented with EWC, it is not only able to achieve similar performance on the new task, but also capable of overcoming a substantial amount of catastrophic forgetting on the two previous tasks.

The existing image feature extraction methods are primarily based on the content and structure information of images, and rarely consider the contextual semantic information. Regarding some types of images such as scenes and objects, the annotations and descriptions of them available on the web may provide reliable contextual semantic information for feature extraction. In this paper, we introduce novel semantic features of an image based on the annotations and descriptions of its similar images available on the web. Specifically, we propose a new method which consists of two consecutive steps to extract our semantic features. For each image in the training set, we initially search the top $k$ most similar images from the internet and extract their annotations/descriptions (e.g., tags or keywords). The annotation information is employed to design a filter bank for each image category and generate filter words (codebook). Finally, each image is represented by the histogram of the occurrences of filter words in all categories. We evaluate the performance of the proposed features in scene image classification on three commonly-used scene image datasets (i.e., MIT-67, Scene15 and Event8). Our method typically produces a lower feature dimension than existing feature extraction methods. Experimental results show that the proposed features generate better classification accuracies than vision based and tag based features, and comparable results to deep learning based features.

Latest addition to the toolbox of human species is Artificial Intelligence(AI). Thus far, AI has made significant progress in low stake low risk scenarios such as playing Go and we are currently in a transition toward medium stake scenarios such as Visual Dialog. In my thesis, I argue that we need to incorporate controlled de-entanglement as first class object to succeed in this transition. I present mathematical analysis from information theory to show that employing stochasticity leads to controlled de-entanglement of relevant factors of variation at various levels. Based on this, I highlight results from initial experiments that depict efficacy of the proposed framework. I conclude this writeup by a roadmap of experiments that show the applicability of this framework to scalability, flexibility and interpretibility.

In this work, we introduce a new approach for the efficient solution of autonomous decision and planning problems, with a special focus on decision making under uncertainty and belief space planning (BSP) in high-dimensional state spaces. Usually, to solve the decision problem, we identify the optimal action, according to some objective function. Instead, we claim that we can sometimes generate and solve an analogous yet simplified decision problem, which can be solved more efficiently. Furthermore, a wise simplification method can lead to the same action selection, or one for which the maximal loss can be guaranteed. This simplification is separated from the state inference, and does not compromise its accuracy, as the selected action would finally be applied on the original state. At first, we develop the concept for general decision problems, and provide a theoretical framework of definitions to allow a coherent discussion. We then practically apply these ideas to BSP problems, in which the problem is simplified by considering a sparse approximation of the initial belief. The scalable sparsification algorithm we provide is able to yield solutions which are guaranteed to be consistent with the original problem. We demonstrate the benefits of the approach in the solution of a highly realistic active-SLAM problem, and manage to significantly reduce computation time, with practically no loss in the quality of solution. This rigorous and fundamental work is conceptually novel, and holds numerous possible extensions.

People are generally afraid of AI, which is using computational statistics to make predictions. "71% of consumers fear AI will infringe on their privacy." A survey of Americans' thoughts on the impact of AI conducted by Oxford concluded this: "34 percent of respondents thought it would be negative, with 12 percent going for the option'very bad, possibly human extinction.'" Further, 18% were uncertain of the impact of AI, meaning that 64% of people have a negative or uncertain view of AI. However, as reasonable as some of these concerns might be, they're more reminiscent of fear of the unknown and the new, rather than problems with AI itself.

It found the impact on thinking skills, memory and behaviour was worse than that of teenage drinking. The researchers, from the University of Montreal, urged teenagers to delay their use of cannabis for as long as they felt able. The study tracked and tested 3,800 adolescents over four years, starting from around the age of 13. Drinking alcohol and taking drugs, such as cannabis, at a young age is known to cause problems with cognitive abilities such as learning, attention and decision-making as well as academic performance at school. This study found these problems increased as cannabis use increased - and the effects were lasting, unlike those of alcohol.

As AI algorithms play a bigger role in decision making, how do qualities like ethics, compassion, and inclusion get programmed into the code? On this edition of Bytemarks Café, a talk about the gathering of thought leaders in Hawai'i to discuss how to move the technology agenda. The event is called TechForce 2019, and its aim is to bring together leaders from key sectors to accelerate tech readiness in our islands. On this edition of Bytemarks Café, a discussion about a novel new project that projects a 3D hologram from Hawaii to American Samoa. The project is called Holo Campus, and is the delivery of University of Hawai'i lectures over the trans-Pacific fiber optic broadband network to the Pacific Islands.

Anadolu Agency called on other European news media organizations to be more sensitive towards the ongoing tragedy in Syria. A three-day general assembly of the European Alliance of News Agencies (EANA) came to an end on Friday in the Czech capital Prague. Anadolu Agency editor-in-chief Metin Mutanoglu said in a speech that Syria's northwestern Idlib area was under heavy fire by Bashar al-Assad regime forces and that the region was facing a fresh wave of migrants. Mutanoglu underlined that though tens of thousands were forced to leave their homes due to regime attacks, the European news media were not interested enough in the issue. A new migration wave would affect not only Turkey but the rest of Europe as well, he stressed, adding that EANA should thus make a greater effort to draw attention to the humanitarian crisis in war-torn country .

Companies have learned the hard way that their artificial intelligence tools have unforeseen outputs, like Amazon's (AMZN) favoring men's resumes over women's or Uber's disabling the user accounts of transgender drivers. When not astutely overseen by human intelligence, deploying AI can often bend into an unseemly rainbow of discriminatory qualities like ageism, sexism, and racism. That's because biases unnoticed in the input data can become amplified in the outputs. Another underappreciated hazard is the potential for AI to cater to our established preferences. You can see that in apps that manage everything from sources of journalism to new music and prospective romance.

As a young child, I used to imagine what life would be like if I had a chauffeur to drive me wherever and whenever I wanted. Of course, this was a luxury afforded by only the wealthy and remained well out of reach for most people -- myself included. Fast forward to today, and the rise of the ride-sharing economy has essentially leveled the playing field, giving everyone affordable access to on-demand transportation. Access to artificial intelligence (AI) is poised to undergo a similar shift. Traditionally, large corporations and government entities have been ahead of the adoption curve because they've had the capital to invest in and the talent to leverage the technology.