Deep neural networks are known to suffer the catastrophic forgetting problem, where they tend to forget the knowledge from the previous tasks when sequentially learning new tasks. Such failure hinders the application of deep learning based vision system in continual learning settings. In this work, we present a simple yet surprisingly effective way of preventing catastrophic forgetting. Our method, called Few-shot Self Reminder (FSR), regularizes the neural net from changing its learned behaviour by performing logit matching on selected samples kept in episodic memory from the old tasks. Surprisingly, this simplistic approach only requires to retrain a small amount of data in order to outperform previous methods in knowledge retention. We demonstrate the superiority of our method to the previous ones in two different continual learning settings on popular benchmarks, as well as a new continual learning problem where tasks are designed to be more dissimilar.

The softmax function has been primarily employed in reinforcement learning (RL) to improve exploration and provide a differentiable approximation to the max function, as also observed in the mellowmax paper by Asadi and Littman. This paper instead focuses on using the softmax function in the Bellman updates, independent of the exploration strategy. Our main theory provides a performance bound for the softmax Bellman operator, and shows it converges to the standard Bellman operator exponentially fast in the inverse temperature parameter. We also prove that under certain conditions, the softmax operator can reduce the overestimation error and the gradient noise. A detailed comparison among different Bellman operators is then presented to show the trade-off when selecting them. We apply the softmax operator to deep RL by combining it with the deep Q-network (DQN) and double DQN algorithms in an off-policy fashion, and demonstrate that these variants can often achieve better performance in several Atari games, and compare favorably to their mellowmax counterpart.

In lifelong learning, the learner is presented with a sequence of tasks, incrementally building a data-driven prior which may be leveraged to speed up learning of a new task. In this work, we investigate the efficiency of current lifelong approaches, in terms of sample complexity, computational and memory cost. Towards this end, we first introduce a new and a more realistic evaluation protocol, whereby learners observe each example only once and hyper-parameter selection is done on a small and disjoint set of tasks, which is not used for the actual learning experience and evaluation. Second, we introduce a new metric measuring how quickly a learner acquires a new skill. Third, we propose an improved version of GEM (Lopez-Paz & Ranzato, 2017), dubbed Averaged GEM (A-GEM), which enjoys the same or even better performance as GEM, while being almost as computationally and memory efficient as EWC (Kirkpatrick et al., 2016) and other regularization-based methods. Finally, we show that all algorithms including A-GEM can learn even more quickly if they are provided with task descriptors specifying the classification tasks under consideration. Our experiments on several standard lifelong learning benchmarks demonstrate that A-GEM has the best trade-off between accuracy and efficiency.

Lack of performance when it comes to continual learning over non-stationary distributions of data remains a major challenge in scaling neural network learning to more human realistic settings. In this work we propose a new conceptualization of the continual learning problem in terms of a temporally symmetric trade-off between transfer and interference that can be optimized by enforcing gradient alignment across examples. We then propose a new algorithm, Meta-Experience Replay (MER), that directly exploits this view by combining experience replay with optimization based meta-learning. This method learns parameters that make interference based on future gradients less likely and transfer based on future gradients more likely. We conduct experiments across continual lifelong supervised learning benchmarks and non-stationary reinforcement learning environments demonstrating that our approach consistently outperforms recently proposed baselines for continual learning. Our experiments show that the gap between the performance of MER and baseline algorithms grows both as the environment gets more non-stationary and as the fraction of the total experiences stored gets smaller.

With the machine learning market size expected to grow from $1.03 Billion USD in 2016 to $8.81 Billion USD by 2022, it can almost be said that machine learning is taking over the world. With that, there is a growing need for professionals who know the ins and out of machine learning. According to Forbes, machine learning patents grew at a 34 percent Compound Annual Growth Rate (CAGR) between 2013 and 2017, which is the third-fastest growing category of all patents granted. Also, the International Data Corporation (IDC) forecasts that spending on AI and ML will increase from $12 Billion USD in 2017 to $57.6 Billion USD by 2021. Even Deloitte Global predicts that the number of machine learning pilots and implementations will double in 2018 compared to 2017, and double again by 2020.

The Air Force is seeking emerging technologies and ways to apply artificial intelligence and cloud-based systems to "enhance the effectiveness" of air education and training, according to a request for information submitted Nov. 27. Air Education and Training Command (AETC) is seeking new methods and processes as well as new products that can "maximize and accelerate learning, both holistically for all Airmen and for individuals," according to service documents. The command is focused on four key investment areas as part of its 2018 strategic plan, including emerging technologies, games and simulation, experiential learning and big data analytics, the RFI said. The Air Force is looking for advanced technology that can contribute to "abundant computer devices, flexible classroom designs, innovative visual displays, games and simulations, collaborative tools, and mechanisms that both track and assess an Airmen's learning efforts," the document said. It is also interested in ways to exploit artificial intelligence, along with virtual and augmented reality systems for areas such as "intelligent tutors," which can help to gauge a student's strengths and weaknesses in areas of study and present materials accordingly. In terms of gaming and simulation efforts, AETC seeks simulation technologies including image generation, visualization and interoperability to help warfighters "keep pace with operational training demands to maximize mission readiness."

Curriculum learning in reinforcement learning is a training methodology that seeks to speed up learning of a difficult target task, by first training on a series of simpler tasks and transferring the knowledge acquired to the target task. Automatically choosing a sequence of such tasks (i.e. a curriculum) is an open problem that has been the subject of much recent work in this area. In this paper, we build upon a recent method for curriculum design, which formulates the curriculum sequencing problem as a Markov Decision Process. We extend this model to handle multiple transfer learning algorithms, and show for the first time that a curriculum policy over this MDP can be learned from experience. We explore various representations that make this possible, and evaluate our approach by learning curriculum policies for multiple agents in two different domains. The results show that our method produces curricula that can train agents to perform on a target task as fast or faster than existing methods.

In Texas, a company called Alegion is helping disabled veterans take part in the new digital economy. The vets' job: preparing data so that an artificial intelligence (AI) system can learn from it. "There's a whole new industry sprouting on the shoulders of AI," says Alegion CEO Nathaniel Gates in an interview with IEEE Spectrum. When people talk about AI, they're often referring to software that gets very good at a particular task via a technique called deep learning. With this method, AI systems are given vast amounts of labeled data, and as they run through it, they learn to draw conclusions.

Machine-learning engineer was the fastest growing job category in the five years to 2017, according to LinkedIn. But tech's hottest role isn't a simple field to break into, requiring at least high school math and some programming knowledge, even to get started. Luckily there are an increasing number of options for those wanting to get a grounding in the field, with Amazon Web Services (AWS) being the latest tech giant to release a set of machine-learning courses for free. That's in addition to the existing well-regarded material available online from the likes of fast.ai and Andrew Ng and Coursera. If you're interested in these courses, it's worth noting that you'll benefit more if you have a basic knowledge of Python and high school linear algebra, statistics, and calculus.

Are you looking for the Best IT Training Online? Grab this Best IT Courses Online & Tutorial which will help you to get the Best IT Certifications Online to skyrocket your career. Information Technology Certifications will assist you to understand the real-life implementation of Artificial Intelligence (AI), Data Analytics and Cloud Computing how this has changed the way we work and the way we think. Taking these Online IT Training 2018-19 will assist you to gain robust knowledge in IT sector and new doors will open for you too. Revolutionary changes have taken places in the IT sector due to some big companies like Space X, Amazon, eBay, Microsoft, Facebook and so on.