Both brain science and the deep learning communities have the problem of interpreting neural activity. For deep learning, even though we can access all neurons' activity data, interpretation of how the deep network solves the task is still challenging. Although a large amount of effort has been devoted to interpreting a deep network, there is still no consensus of what interpretation is. This paper tries to push the discussion in this direction and proposes an information-theoretic progressive framework to synthesize interpretation. Firstly, we discuss intuitions of interpretation: interpretation is meta-information; interpretation should be at the right level; inducing independence is helpful to interpretation; interpretation is naturally progressive; interpretation doesn't have to involve a human. Then, we build the framework with an information map splitting idea and implement it with the variational information bottleneck technique. After that, we test the framework with the CLEVR dataset. The framework is shown to be able to split information maps and synthesize interpretation in the form of meta-information.

We propose a method for meta-learning reinforcement learning algorithms by searching over the space of computational graphs which compute the loss function for a value-based model-free RL agent to optimize. The learned algorithms are domain-agnostic and can generalize to new environments not seen during training. Our method can both learn from scratch and bootstrap off known existing algorithms, like DQN, enabling interpretable modifications which improve performance. Bootstrapped from DQN, we highlight two learned algorithms which obtain good generalization performance over other classical control tasks, gridworld type tasks, and Atari games. The analysis of the learned algorithm behavior shows resemblance to recently proposed RL algorithms that address overestimation in value-based methods. Designing new deep reinforcement learning algorithms that can efficiently solve across a wide variety of problems generally requires a tremendous amount of manual effort. Learning to design reinforcement learning algorithms or even small sub-components of algorithms would help ease this burden and could result in better algorithms than researchers could design manually. Our work might then shift from designing these algorithms manually into designing the language and optimization methods for developing these algorithms automatically. Reinforcement learning algorithms can be viewed as a procedure that maps an agent's experience to a policy that obtains high cumulative reward over the course of training. We formulate the problem of training an agent as one of meta-learning: an outer loop searches over the space of computational graphs or programs that compute the objective function for the agent to minimize and an inner loop performs the updates using the learned loss function. The objective of the outer loop is to maximize the training return of the inner loop algorithm.

Artificial Intelligence (AI) is a discipline of computer science that deals with machine intelligence. It is essential to bring AI into the context of learning because it helps in analysing the enormous amounts of data that is collected from individual students, teachers and academic staff. The major priorities of implementing AI in education are making innovative use of existing digital technologies for learning, and teaching practices that significantly improve traditional educational methods. The main problem with traditional learning is that it cannot be suited to every student in class. Some students may grasp the concepts well, while some may have difficulties in understanding them and some may be more auditory or visual learners. The World Bank report on education has indicated that the learning gap created by this problem causes many students to drop out (World Development Report, 2018). Personalised learning has been able to solve this grave problem.

Continual relation extraction is an important task that focuses on extracting new facts incrementally from unstructured text. Given the sequential arrival order of the relations, this task is prone to two serious challenges, namely catastrophic forgetting and order-sensitivity. We propose a novel curriculum-meta learning method to tackle the above two challenges in continual relation extraction. We combine meta learning and curriculum learning to quickly adapt model parameters to a new task and to reduce interference of previously seen tasks on the current task. We design a novel relation representation learning method through the distribution of domain and range types of relations. Such representations are utilized to quantify the difficulty of tasks for the construction of curricula. Moreover, we also present novel difficulty-based metrics to quantitatively measure the extent of order-sensitivity of a given model, suggesting new ways to evaluate model robustness. Our comprehensive experiments on three benchmark datasets show that our proposed method outperforms the state-of-the-art techniques. The code is available at the anonymous GitHub repository: https://github.com/wutong8023/AAAI_CML.

Kimari Rennis, an 18-year-old who grew up in the Bronx, first became involved with the NYVGCC through an after-school program at DreamYard Prepatory School. Upon completing the after-school program four years ago, she became a paid intern, and is still in the internship program today. She calls the program "very generous" and "life changing," an opportunity to hone her journalistic skills and also meet luminaries within the video games industry, such as Hideo Kojima and actor Norman Reedus. Eventually, she received a scholarship through the NYVGCC, granting her the ability to enroll in the NYU, where she is currently studying game design.

In a digital, global world where information is projected to double every 12 hours ([ 1 ][1]), the memorization of facts will become less of a commodity than the ability to think, find patterns, and generate new ideas from old parts ([ 2 ][2], [ 3 ][3]). Thus, a cradle-to-career approach to educating children must be mindful of how children learn to learn, not just what they learn ([ 4 ][4]). Combining insight, scientific acumen, and exquisite narrative, The Intellectual Lives of Children allows readers to peer into the minds of infants, toddlers, and preschoolers as they explore and learn in everyday moments, emphasizing what constitutes real learning. Children are bursting with playful curiosity. By age 3, they ask questions about everything they see—Why does a tree have leaves? Why does the Sun come up each day?—and by age 5, they pose even deeper questions, about God and morals. These questions not only provide fodder for knowledge, they help children discover the causal relationships among things—all with adult mentors by their side. Children also need time to explore. One child might collect dead things like worms and slugs, and another, assorted leaves of different shapes and colors. These collections, Engel argues, become treasured resources for the discovery of patterns, and they invite even more inquisitiveness. Indeed, the adults who guide this exploration by asking questions themselves reinforce curiosity and innovation. Hidden in these playful encounters are rich opportunities for learning. Yet explorations take time—the time to meander and discover, the unscheduled time to be bored. As Engel writes, “when children are allowed to dive into a topic thoroughly, they…connect isolated facts in order to generate new ideas.” They learn grit and they learn to have agency over their own learning. As such, the real mental work for children takes place in plain sight as they play—when a child builds a platform of chairs and pillows to retrieve cookies from an out-of-reach cookie jar and when she uses kitchen utensils to fish for the toy that is lodged under the couch. As adults, we often overlook the fact that learning is happening during periods of unstructured play, or we dismiss these intervals as unproductive. Hurried parents often lack the ability to carve out that time, fearing that their children might be late for their next scheduled activity. “Watch and listen for twenty minutes in almost any school in the United States and it becomes clear that the educational system does not concern itself with children's intellectual lives,” admonishes Engel in the opening pages of the book. Instead, she hopes to reenvision schools as “idea factories” built on inspiring curiosity and problem solving: “Imagine assessing students' progress under some new headings: poses interesting questions, speculates,…articulates important problems and spends time solving them.” In one lovely example, Engel describes a teacher who challenged her students to construct a record-breaking straw chain that would eventually measure 3.8 miles. “Winning the record would be fun, but the enduring benefit would be coming to grips with vast quantities,” explains the teacher, whose goal was to help the children to better understand the sheer depth of the Mariana Trench. The puzzles and problems that captivate children and the ways they set about solving them are reminiscent of how philosophers Karl Popper and Thomas Kuhn conceptualized the thinking of scientists ([ 5 ][5], [ 6 ][6]). Both children and scientists bring the tools in their respective arsenals to bear on things that matter to them. Their learning is not linear and is certainly not funneled through flashcards ([ 7 ][7]). In the past few decades, developmental science has made great strides in understanding the mental richness of infants, toddlers, and preschoolers. Engel's book helps parents and educators see what scientists have learned, offering tips for how to make the learning even more apparent. For example, she encourages parents to see children as active thinkers and suggests that by asking open-ended questions and letting them explore, children will be better prepared to thrive in a complex and ever-changing world. 1. [↵][8]1. S. Sorkin , “Thriving in a world of ‘knowledge half-life’,” Enterprising Insights, 5 April 2019. 2. [↵][9]1. R. M. Golinkoff, 2. K. Hirsh-Pasek , Becoming Brilliant (APA Press, 2016). 3. [↵][10]1. D. H. Pink , A Whole New Mind (Penguin, 2006). 4. [↵][11]1. K. Hirsh-Pasek, 2. H. S. Hadani, 3. E. Blinkoff, 4. R. M. Golinkoff , “A new path to education reform: Playful learning promotes 21st-century skills in schools and beyond,” The Brookings Institution: Big Ideas Policy Report, 28 October 2020. 5. [↵][12]1. K. Popper , The Logic of Scientific Discovery (Hutchinson, 1959). 6. [↵][13]1. T. S. Kuhn , The Structure of Scientific Revolutions (Univ. of Chicago Press, 1962). 7. [↵][14]1. A. Gopnik, 2. A. N. Meltzoff, 3. P. K. Kuhl , The Scientist in the Crib (William Morrow, 1999). [1]: #ref-1 [2]: #ref-2 [3]: #ref-3 [4]: #ref-4 [5]: #ref-5 [6]: #ref-6 [7]: #ref-7 [8]: #xref-ref-1-1 "View reference 1 in text" [9]: #xref-ref-2-1 "View reference 2 in text" [10]: #xref-ref-3-1 "View reference 3 in text" [11]: #xref-ref-4-1 "View reference 4 in text" [12]: #xref-ref-5-1 "View reference 5 in text" [13]: #xref-ref-6-1 "View reference 6 in text" [14]: #xref-ref-7-1 "View reference 7 in text"

Last month, the British television network Channel 4 broadcast an "alternative Christmas address" by Queen Elizabeth II, in which the 94-year-old monarch was shown cracking jokes and performing a dance popular on TikTok. Of course, it wasn't real: The video was produced as a warning about deepfakes--apparently real images or videos that show people doing or saying things they never did or said. If an image of a person can be found, new technologies using artificial intelligence and machine learning now make it possible to show that person doing almost anything at all. The dangers of the technology are clear: A high-school teacher could be shown in a compromising situation with a student, a neighbor could be depicted as a terrorist. Can deepfakes, as such, be prohibited under American law?

Working with a non-stationary stream of data requires for the analysis system to evolve its model (the parameters as well as the structure) over time. In particular, concept drifts can occur, which makes it necessary to forget knowledge that has become obsolete. However, the forgetting is subjected to the stability-plasticity dilemma, that is, increasing forgetting improve reactivity of adapting to the new data while reducing the robustness of the system. Based on a set of inference rules, Evolving Fuzzy Systems-EFS-have proven to be effective in solving the data stream learning problem. However tackling the stability-plasticity dilemma is still an open question. This paper proposes a coherent method to integrate forgetting in Evolving Fuzzy System, based on the recently introduced notion of concept drift anticipation. The forgetting is applied with two methods: an exponential forgetting of the premise part and a deferred directional forgetting of the conclusion part of EFS to preserve the coherence between both parts. The originality of the approach consists in applying the forgetting only in the anticipation module and in keeping the EFS (called principal system) learned without any forgetting. Then, when a drift is detected in the stream, a selection mechanism is proposed to replace the obsolete parameters of the principal system with more suitable parameters of the anticipation module. An evaluation of the proposed methods is carried out on benchmark online datasets, with a comparison with state-of-the-art online classifiers (Learn++.NSE, PENsemble, pclass) as well as with the original system using different forgetting strategies.

Understanding human language is one of the key themes of artificial intelligence. For language representation, the capacity of effectively modeling the linguistic knowledge from the detail-riddled and lengthy texts and getting rid of the noises is essential to improve its performance. Traditional attentive models attend to all words without explicit constraint, which results in inaccurate concentration on some dispensable words. In this work, we propose using syntax to guide the text modeling by incorporating explicit syntactic constraints into attention mechanisms for better linguistically motivated word representations. In detail, for self-attention network (SAN) sponsored Transformer-based encoder, we introduce syntactic dependency of interest (SDOI) design into the SAN to form an SDOI-SAN with syntax-guided self-attention. Syntax-guided network (SG-Net) is then composed of this extra SDOI-SAN and the SAN from the original Transformer encoder through a dual contextual architecture for better linguistics inspired representation. The proposed SG-Net is applied to typical Transformer encoders. Extensive experiments on popular benchmark tasks, including machine reading comprehension, natural language inference, and neural machine translation show the effectiveness of the proposed SG-Net design.

Transfer learning seeks to improve the generalization performance of a target task by exploiting the knowledge learned from a related source task. Central questions include deciding what information one should transfer and when transfer can be beneficial. The latter question is related to the so-called negative transfer phenomenon, where the transferred source information actually reduces the generalization performance of the target task. This happens when the two tasks are sufficiently dissimilar. In this paper, we present a theoretical analysis of transfer learning by studying a pair of related perceptron learning tasks. Despite the simplicity of our model, it reproduces several key phenomena observed in practice. Specifically, our asymptotic analysis reveals a phase transition from negative transfer to positive transfer as the similarity of the two tasks moves past a well-defined threshold. Transfer learning [1]-[5] is a promising approach to improving the performance of machine learning tasks. It does so by exploiting the knowledge gained from a previously-learned model, referred to as the source task, to improve the generalization performance of a related learning problem, referred to as the target task.