What if I told a story here, how would that story start?" Thus, the summarization prompt: "My second grader asked me what this passage means: …" When a given prompt isn't working and GPT-3 keeps pivoting into other modes of completion, that may mean that one hasn't constrained it enough by imitating a correct output, and one needs to go further; writing the first few words or sentence of the target output may be necessary.
The latest Terminator movie, Dark Fate, struggles to give satisfying emotional arcs to its large cast of characters. Writer Sara Lynn Michener says it doesn't help that a large chunk of the movie is wasted on a bombastic action sequence set aboard an exploding cargo plane. "I think there's this idea with, especially, male directors where they get really excited about trying to top what's been done before, but do it even bigger and better and more Michael Bay-ish," Michener says in Episode 386 of the Geek's Guide to the Galaxy podcast. Are we really doing that in 2019? Geek's Guide to the Galaxy host David Barr Kirtley agrees that the cargo plane sequence was silly, and stands in sharp contrast to the sense of realism captured in the franchise's best installments, The Terminator and Terminator 2: Judgment Day.
We propose a novel neural topic model in the Wasserstein autoencoders (W AE) framework. Unlike existing variational autoencoder based models, we directly enforce Dirichlet prior on the latent document-topic vectors. We exploit the structure of the latent space and apply a suitable kernel in minimizing the Maximum Mean Discrepancy (MMD) to perform distribution matching. We discover that MMD performs much better than the Generative Adversarial Network (GAN) in matching high dimensional Dirichlet distribution. We further discover that incorporating randomness in the encoder output during training leads to significantly more coherent topics. To measure the diversity of the produced topics, we propose a simple topic uniqueness metric. Together with the widely used coherence measure NPMI, we offer a more wholistic evaluation of topic quality. Experiments on several real datasets show that our model produces significantly better topics than existing topic models.
With the advent of deep neural networks, some research focuses towards understanding their black-box behavior. In this paper, we propose a new type of self-interpretable models, that are, architectures designed to provide explanations along with their predictions. Our method proceeds in two stages and is trained end-to-end: first, our model builds a low-dimensional binary representation of any input where each feature denotes the presence or absence of concepts. Then, it computes a prediction only based on this binary representation through a simple linear model. This allows an easy interpretation of the model's output in terms of presence of particular concepts in the input. The originality of our approach lies in the fact that concepts are automatically discovered at training time, without the need for additional supervision. Concepts correspond to a set of patterns, built on local low-level features (e.g a part of an image, a word in a sentence), easily identifiable from the other concepts. We experimentally demonstrate the relevance of our approach using classification tasks on two types of data, text and image, by showing its predictive performance and interpretability.
While many methods for interpreting machine learning models have been proposed, they are frequently ad hoc, difficult to evaluate, and come with no statistical guarantees on the error rate. This is especially problematic in scientific domains, where interpretations must be accurate and reliable. In this paper, we cast black box model interpretation as a hypothesis testing problem. The task is to discover "important" features by testing whether the model prediction is significantly different from what would be expected if the features were replaced with randomly-sampled counterfactuals. We derive a multiple hypothesis testing framework for finding important features that enables control over the false discovery rate. We propose two testing methods, as well as analogs of one-sided and two-sided tests. In simulation, the methods have high power and compare favorably against existing interpretability methods. When applied to vision and language models, the framework selects features that intuitively explain model predictions.
And even these engagement patterns are giving way to new and more seamless and natural methods of interaction. For example, images and video feeds can be used to track assets, authenticate individual identities, and understand context from surrounding environments. Advanced voice capabilities allow interaction with complex systems in natural, nuanced conversations. Moreover, by intuiting human gestures, head movements, and gazes, AI-based systems can respond to nonverbal user commands. Intelligent interfaces combine the latest in human-centered design techniques with leading-edge technologies such as computer vision, conversational voice, auditory analytics, and advanced augmented reality and virtual reality. Working in concert, these techniques and capabilities are transforming the way we engage with machines, data, and each other. At a dinner party, your spouse, across the table, raises an eyebrow ever so slightly. The gesture is so subtle that no one else notices, but you received the message loud and clear: "I'm bored. Most people recognize this kind of intuitive communication as a shared language that develops over time among people in intimate relationships. We accept it as perfectly natural--but only between humans.
We'll wish a Happy Thanksgiving to those celebrating and a merry Thursday to those who are not. Google has a movie promo you should probably check out, and have you heard about how MIT's plane flies without any moving parts? Even if it's the cutest phone we've seen in years.Palm phone review: A tiny'second phone' no one needs Unfortunately, the Palm's size will make it too difficult for some to use, while some compatibility issues and poorly executed features are sure to frustrate others. Palm's ambitions were admirable, but as Chris Velazco explains, this $350 device feels like a half-baked answer to a serious problem. Cyber Week has begun.Google makes all movie rentals just 99 cents for Thanksgiving If you've got four days off for Thanksgiving weekend, Google has ways for you to kill time.
Around the time Leonardo Da Vinci was painting the Mona Lisa, he was also writing his Codex on the Flight of Birds, a roughly 35,000-word exploration of the ways in which man might take to the air. His illustrations included diagrams positing pre-Newtonian theories of physics, a rudimentary plan for a flying machine and many, many sketches of birds in flight. The Mona Lisa, with her secretive smile, is a universe of intimacy captured on a relatively small panel of wood. But the landscape behind his captivating subject shows the world as you would see it from atop a tall hill--or from the vantage point you would get if you had hitched a ride on the back of a giant bird. Even as da Vinci was perfecting one way of seeing a face, he was dreaming of other ways of looking.
We present a probabilistic framework for studying adversarial attacks on discrete data. Based on this framework, we derive a perturbation-based method, Greedy Attack, and a scalable learning-based method, Gumbel Attack, that illustrate various tradeoffs in the design of attacks. We demonstrate the effectiveness of these methods using both quantitative metrics and human evaluation on various state-of-the-art models for text classification, including a word-based CNN, a character-based CNN and an LSTM. As as example of our results, we show that the accuracy of character-based convolutional networks drops to the level of random selection by modifying only five characters through Greedy Attack.