Feature matters for salient object detection. Existing methods mainly focus on designing a sophisticated structure to incorporate multi-level features and filter out cluttered features. We present Progressive Feature Polishing Network (PFPN), a simple yet effective framework to progressively polish the multi-level features to be more accurate and representative. By employing multiple Feature Polishing Modules (FPMs) in a recurrent manner, our approach is able to detect salient objects with fine details without any post-processing. A FPM parallelly updates the features of each level by directly incorporating all higher level context information. Moreover, it can keep the dimensions and hierarchical structures of the feature maps, which makes it flexible to be integrated with any CNN-based models. Empirical experiments show that our results are monotonically getting better with increasing number of FPMs. Without bells and whistles, PFPN outperforms the state-of-the-art methods significantly on five benchmark datasets under various evaluation metrics.

At the Common Good in the Digital Age tech conference recently held in Vatican City, Pope Francis urged Facebook executives, venture capitalists, and government regulators to be wary of the impact of AI and other technologies. "If mankind's so-called technological progress were to become an enemy of the common good, this would lead to an unfortunate regression to a form of barbarism dictated by the law of the strongest," he said. In a related but contextually different conversation, this summer Joy Buolamwini testified before Congress with Rep. Alexandria Ocasio-Cortez (D-NY) that multiple audits found facial recognition technology generally works best on white men and worst on women of color. What these two events have in common is their relationship to power dynamics in the AI ethics debate. Arguments about AI ethics can wage without mention of the word "power," but it's often there just under the surface. In fact, it's rarely the direct focus, but it needs to be. Power in AI is like gravity, an invisible force that influences every consideration of ethics in artificial intelligence. Power provides the means to influence which use cases are relevant; which problems are priorities; and who the tools, products, and services are made to serve. It underlies debates about how corporations and countries create policy governing use of the technology.

Recent methods for training generalized additive models (GAMs) with pairwise interactions achieve state-of-the-art accuracy on a variety of datasets. Adding interactions to GAMs, however, introduces an identifiability problem: effects can be freely moved between main effects and interaction effects without changing the model predictions. In some cases, this can lead to contradictory interpretations of the same underlying function. This is a critical problem because a central motivation of GAMs is model interpretability. In this paper, we use the Functional ANOV A decomposition to uniquely define interaction effects and thus produce identifiable additive models with purified interactions. To compute this decomposition, we present a fast, exact, mass-moving algorithm that transforms any piecewise-constant function (such as a tree-based model) into a purified, canonical representation. We apply this algorithm to several datasets and show large disparity, including contradictions, between the apparent and the purified effects. An important question in data analysis is whether two variables act in concert to affect an outcome. But this unconstrained additive model has fundamental flaws.

Farmland in Fukushima that was rendered unusable after the disastrous 2011 nuclear meltdown is getting a second chance at productivity. A group of Japanese investors have created a new plan to use the abandoned land to build wind and solar power plants, to be used to send electricity to Tokyo. The plan calls for the construction of eleven solar power plants and ten wind power plants, at an estimated cost of $2.75 billion. Fukushima has been aggressively converting land damaged by the 2011 meltdown, such as this golf course (pictured above) into a source of renewable energy. A new $2.75 billion plan will add eleven new solar plants and ten wind power plants to former farmland The project is expected to be completed in March of 2024 and is backed by a group of investors, including Development Bank of Japan and Mizuho Bank.

The growing demand for ML/AI technologies, as well as for ML/AI talent, in the pharmaceutical industry is driving the formation of a new interdisciplinary field: data-driven drug discovery/healthcare. Consequently, there is a growing number of AI driven startups offering technology solutions for drug discovery/development. In drug development, preclinical phase (in vitro and in vivo), also named preclinical studies and nonclinical studies, is a stage of research that begins before clinical trials, and during which important feasibility, iterative testing and drug safety data are collected. According to a detailed mind-map prepared by Pharma Division of Deep Knowledge Analytics (updated Q1 2019): the AI for Drug Discovery, Biomarker Development and Advanced R&D Industry Landscape counts so far 400 investors, 170 companies and 50 corporations. This article focuses only on the AI startups and the AI investors trying to overcome the above 4 challenges during design and execution of the preclinical phase.

Visual 1st, the annual Silicon-Valley imaging conference for industry leaders and upstarts, once again brought together a worldwide audience for a day-and-a-half executive conference. The event, held Oct. 2-3 at the Golden Gate Club in San Francisco, addresses topics as far-reaching as artificial intelligence and as every day as printing. As with most conferences, the real meat of the event is the hallway discussions and informal meetings over a beer or wine at the reception. Below are some photos from the conference, courtesy of sponsor, Sweet Escapes. Each year, a panel of high-powered industry experts presented the four Visual 1st Awards to the most outstanding among 30 products competing in this year's show-and-tell demo sessions.

Alexa, Google Assistant, Siri, and Cortana can answer all sorts of questions that pop into users' heads, and they're improving every day. But what happens when a company like Amazon decides to crowdsource answers to fill gaps in its platform's knowledge? The result can range from amusing and perplexing to concerning. Alexa Answers allows any Amazon customer to submit responses to unanswered questions. When the web service launched in general availability a few weeks ago, Amazon gave assurances that submissions would be policed through a combination of automatic and manual review.

Learning with rejection (LWR) allows development of machine learning systems with the ability to discard low confidence decisions generated by a prediction model. That is, just like human experts, LWR allows machine models to abstain from generating a prediction when reliability of the prediction is expected to be low. Several frameworks for this learning with rejection have been proposed in the literature. However, most of them work for classification problems only and regression with rejection has not been studied in much detail. In this work, we present a neural framework for LWR based on a generalized meta-loss function that involves simultaneous training of two neural network models: a predictor model for generating predictions and a rejecter model for deciding whether the prediction should be accepted or rejected. The proposed framework can be used for classification as well as regression and other related machine learning tasks. We have demonstrated the applicability and effectiveness of the method on synthetically generated data as well as benchmark datasets from UCI machine learning repository for both classification and regression problems. Despite being simpler in implementation, the proposed scheme for learning with rejection has shown to perform at par or better than previously proposed methods. Furthermore, we have applied the method to the problem of hurricane intensity prediction from satellite imagery. Significant improvement in performance as compared to conventional supervised methods shows the effectiveness of the proposed scheme in real-world regression problems.

New generation geostationary satellites make solar reflectance observations available at a continental scale with unprecedented spatiotemporal resolution and spectral range. Generating quality land monitoring products requires correction of the effects of atmospheric scattering and absorption, which vary in time and space according to geometry and atmospheric composition. Many atmospheric radiative transfer models, including that of Multi-Angle Implementation of Atmospheric Correction (MAIAC), are too computationally complex to be run in real time, and rely on precomputed look-up tables. Additionally, uncertainty in measurements and models for remote sensing receives insufficient attention, in part due to the difficulty of obtaining sufficient ground measurements. In this paper, we present an adaptation of Bayesian Deep Learning (BDL) to emulation of the MAIAC atmospheric correction algorithm. Emulation approaches learn a statistical model as an efficient approximation of a physical model, while machine learning methods have demonstrated performance in extracting spatial features and learning complex, nonlinear mappings. We demonstrate stable surface reflectance retrieval by emulation (R2 between MAIAC and emulator SR are 0.63, 0.75, 0.86, 0.84, 0.95, and 0.91 for Blue, Green, Red, NIR, SWIR1, and SWIR2 bands, respectively), accurate cloud detection (86\%), and well-calibrated, geolocated uncertainty estimates. Our results support BDL-based emulation as an accurate and efficient (up to 6x speedup) method for approximation atmospheric correction, where built-in uncertainty estimates stand to open new opportunities for model assessment and support informed use of SR-derived quantities in multiple domains.

Artificial intelligence might sound like a futuristic concept, and it may be true that we're years or decades away from a generalized form of AI that can match or exceed the capabilities of the human brain across a wide range of topics. But the implications of machine learning, facial recognition and other early forms of the technology are already playing out for companies, governmental agencies and people around the world. This is raising questions about everything from privacy to jobs to law enforcement to the future of humanity. On this episode of the GeekWire Podcast, we hear several different takes from people grappling right now with AI and its implications for business, technology and society, recorded across different sessions at the recent GeekWire Summit in Seattle. Listen to the episode above, or subscribe in your favorite podcast app, and continue reading for edited excerpts. Smith: I think it's fair to say that artificial intelligence will reshape the global economy over the next three decades probably more than any other single technological force, probably as much as the combustion engine reshaped the global economy in the first half of the 20th century. One of our chapters is about AI in the workforce, and we actually start it by talking about the role of horses, the last run of the fire of horses in Brooklyn in 1922. And we trace how the transition from the horse to the automobile changed every aspect of the economy. I think the same thing will be true of AI, so we should get that right.