We make three contributions toward better understanding policy gradient methods in the tabular setting. First, we show that with the true gradient, policy gradient with a softmax parametrization converges at a $O(1/t)$ rate, with constants depending on the problem and initialization. This result significantly expands the recent asymptotic convergence results. The analysis relies on two findings: that the softmax policy gradient satisfies a \L{}ojasiewicz inequality, and the minimum probability of an optimal action during optimization can be bounded in terms of its initial value. Second, we analyze entropy regularized policy gradient and show that it enjoys a significantly faster linear convergence rate $O(e^{-t})$ toward softmax optimal policy. This result resolves an open question in the recent literature. Finally, combining the above two results and additional new $\Omega(1/t)$ lower bound results, we explain how entropy regularization improves policy optimization, even with the true gradient, from the perspective of convergence rate. The separation of rates is further explained using the notion of non-uniform \L{}ojasiewicz degree. These results provide a theoretical understanding of the impact of entropy and corroborate existing empirical studies.

SCI COMMUN### Computer science The United States will establish a dozen centers to study artificial intelligence (AI) and quantum information science (QIS), the White House announced last week. The seven university-based AI centers will receive $20 million each over 5 years from the National Science Foundation or the Department of Agriculture and will use AI—algorithms that can learn to recognize patterns—to tackle problems in areas ranging from farming to particle physics. The five centers on QIS, located at the Department of Energy's national laboratories, will focus on topics such as developing quantum computers that could solve challenges that would overwhelm conventional computers. Each of these centers will receive $125 million over 5 years, as Congress called for in the 2018 National Quantum Initiative Act. ### Archaeology Tiny parasitic worms known as helminths cause malnutrition and developmental disorders in some 1.5 billion people around the world, mostly in developing countries. Scientists now report new evidence that better sanitation can alleviate this scourge. During the Middle Ages and for centuries after, worm infections were as prevalent among Europeans as they are today in people living in parts of sub-Saharan Africa, South America, and East Asia, the authors report in a paper published on 27 August in PLOS Neglected Tropical Disease . They based the conclusion on an analysis of 589 samples from skeletons in medieval cemeteries in the Czech Republic, Germany, and the United Kingdom. Because such worms were eradicated in Europe before effective antiparasite drugs were developed, the results reinforce the idea that improvements to water supplies, sanitation, and hygiene can dramatically reduce the disease burden they cause today. ### Disasters When it roared ashore last week in Louisiana, Hurricane Laura packed a double whammy, endangering public safety with its wind and water and slowing efforts to stem the COVID-19 pandemic. Its top wind speed at landfall, 241 kilometers per hour, was the fifth highest documented for any U.S. hurricane. Laura tied a record for the fastest intensifying storm in the Gulf of Mexico, with its wind increasing on 26 August by 105 kilometers per hour in just 24 hours; the causes of such rapid strengthening are little understood. The storm led to at least 19 deaths in Louisiana and Texas. It also threatened to accelerate the spread of COVID-19; testing centers were temporarily closed, and residents of southwest Louisiana, which bore the storm's brunt and had been recording some of the state's highest rates of positive test results, evacuated elsewhere. Seven hurricanes and tropical storms have hit the United States so far this year, one of the most active seasons on record. ### Agriculture The Dutch government last week decided to end mink farming to prevent the animals from becoming sources of the virus that causes COVID-19. More than 40 mink farms in the Netherlands—almost one in three—have had outbreaks of the virus since late April, triggering massive culls. A Dutch law adopted in 2012 banned mink farming by 2024 for ethical reasons, but now the remaining farms must close by March 2021. The government has set aside €182 million to indemnify farmers. Although farms implemented hygiene rules, scientists suspect infected people carried the virus into them. Denmark, Spain, and the United States have seen outbreaks at mink farms as well. ### Public health By testing dormitory wastewater for SARS-CoV-2, the virus that causes COVID-19, the University of Arizona may have stamped out a potential outbreak before it could spread. Several countries, U.S. municipalities, and some universities have been checking sewage for RNA from the virus, which can signal infections shortly before clinical cases and deaths are recorded. In Arizona, officials announced last week that wastewater from a student dormitory contained the viral RNA just days after students had moved into their rooms in August; all 311 residents and dorm workers had previously tested negative on a mandatory test for COVID-19. The university retested all of them and found two students who were asymptomatic but positive for the virus; they were then quarantined. ### Diagnostics The U.S. Food and Drug Administration last week issued an emergency use authorization to Abbott, a laboratory company, for a 15-minute test for the COVID-19 virus that could help expand the number of Americans regularly tested. The new diagnostic, called BinaxNOW, detects proteins, or antigens, that are unique to the virus with high accuracy and at a cost of only $5 each. Other coronavirus tests that identify genetic material unique to the virus typically cost $100, and laboratories often take days to provide results. The genetic tests and other, antigen-based ones require specialized lab equipment; Abbott's does not, although a health care professional must administer it. The company says it plans to produce 50 million tests in October. Last week, the Trump administration announced it would buy 150 million. The United States currently conducts about 700,000 tests for the virus per day. ### Policy The U.S. Centers for Disease Control and Prevention (CDC) drew criticism last week for revising its guidelines to state that people exposed to the virus that causes COVID-19 “do not necessarily need a test” if they lack symptoms and do not have medical conditions that make them vulnerable. Scientists and public health specialists slammed the 24 August revision, noting that people who do not feel sick can still spread the virus and that the United States continues to lead the world in COVID-19 cases and deaths. Trump administration officials have said too many people have been getting tested out of fear and tests should be reserved for those at highest risk, The New York Times reported. But CDC Director Robert Redfield appeared to muddy the message when he said on 26 August that testing “may be considered for all close contacts of confirmed or probable COVID-19 patients.” ### Infectious diseases The EcoHealth Alliance, a nonprofit whose highly scored grant to study bat coronaviruses that could jump to humans in China was summarily defunded after President Donald Trump targeted it, has received new funding worth $7.5 million over 5 years, the U.S. National Institutes of Health (NIH) announced last week. In April, Trump alleged without evidence that the COVID-19 virus escaped from the Wuhan Institute of Virology; the EcoHealth Alliance had collaborated with scientists there on the canceled grant. NIH ended it days later, drawing strong protests from scientists. The newly funded work will not revive the earlier project but instead will focus on risks of animal viruses jumping to humans in Southeast Asia, but not China. The EcoHealth Alliance is one of 11 groups NIH plans to fund with $82 million to study such risks. “It's a relief for us to know that NIH isn't going to blackball our organization because of political interference,” EcoHealth Alliance President Peter Daszak says. ### Funding China continued in 2019 its yearslong run of double-digit annual percentage increases in spending on R&D. But it has not yet reached its long-standing goal of increasing R&D expenditures to 2.5% of gross domestic product (GDP). Total public and private science and technology expenditures in 2019 rose 12.5% to 2.21 trillion Chinese yuan ($322 billion), the National Bureau of Statistics of China reported last week. Most (83%) went to development, while basic research received 6% and applied research 11%. Relative to other countries, China has been spending more on development and less on basic research. Its total R&D spending in 2019 amounted to 2.23% of GDP, still short of the United States's 2.83%. China was the world's second biggest spender on R&D behind the United States in 2018, the latest year for which a comparison is available, according to the Organisation for Economic Co-operation and Development. Analysts expect China to continue to close the gap. ### Astronomy Gravitational wave hunters have netted a big fish: the signal from a pair of black holes merging to produce one with a mass of about 142 Suns. That heft makes it the first confirmed intermediate-mass black hole, with a mass between those produced by collapsing stars and the giant black holes at the hearts of galaxies. Detected in May 2019 by the twin Laser Interferometer Gravitational-wave Observatory facilities in the United States and the Virgo detector in Italy, the merger is also the most distant seen, at 7 billion light-years away, as well as the most powerful, with the mass of eight Suns converted into energy. The masses of the individual black holes—85 and 66 solar masses—before they merged pose a puzzle, as theorists believe it impossible to make a black hole heavier than 65 Suns from the collapse of a single star. The discovery is reported this week in Physical Review Letters and The Astrophysical Journal Letters . ### Nonproliferation After a monthslong impasse, Iran has agreed to allow international inspectors access to two sites that were allegedly part of a clandestine nuclear weapons program. The move preserves, for now, what remains of a multination nuclear deal reached in 2015, from which the Trump administration has withdrawn. The inspections will take place at Abadeh, a testing range for high explosives in central Iran, and at an undisclosed site, which intelligence reports revealed might have contained undeclared nuclear materials and activities. Iran had rebuffed requests from the International Atomic Energy Agency to take samples at the sites; continued stonewalling could have prompted the agency to declare Iran out of compliance with its commitments. The United States maintains that Iran has violated the nuclear deal, and banking and other sanctions lifted after the 2015 accord must automatically resume. But members of the United Nations Security Council last week reiterated their disagreement with that interpretation, and the United States now plans to reimpose those sanctions unilaterally on 20 September. ### Environment The Trump administration on 31 August eased rules on toxic wastewater created by coal-burning power plants, which operators discharge into rivers and streams. The move changes a rule adopted in 2015 by former President Barack Obama's administration requiring plant operators to treat and recycle water used to store coal ash, which contains mercury and arsenic, by 2023. The Trump administration's version instead exempts plants set to close or switch to natural gas by 2028 and allows other plants to delay compliance until that year if they voluntarily adopt advanced biological treatment. The administration says its rule will save the industry money and retain coal-industry jobs while reducing total pollution by 1 million pounds annually, over and above the 1.4 million pound reduction anticipated under the Obama rule. But environmental groups rejected those assertions and predicted that power plants—now the largest contributors of industrial water pollution—will discharge even more. The critics add that the move will prop up coal power, which is responsible for emitting a significant share of global warming gases. ### Conservation A Norwegian wind farm has devised an inexpensive method that may prevent birds from being killed by turbines' rotating blades. By painting only one turbine blade black, the farm reduced bird collisions by more than 70%, say researchers who conducted the first field study of the approach. Fast-moving, monotone blades can be difficult for birds to see; in the United States alone, collisions with wind turbines kill 140,000 to 500,000 birds each year. But a single contrasting black blade makes this rotating obstacle easier for birds to identify and avoid, researchers report in the 27 August issue of Ecology and Evolution . The approach needs further validation, other researchers say. And they note that windmills still rank low on the list of threats to birds: Collisions with power wires and communication towers kill an estimated 32 million birds in the United States annually, for example, and cats are believed to kill 2.4 billion each year. Loss of habitat is another leading threat. ### Infectious diseases Togo is the first African country to have eliminated Human African trypanosomiasis (HAT), better known as sleeping sickness, as a public health problem. The World Health Organization (WHO) on 25 August certified the country as free of HAT, which is caused by two subspecies of the Trypanosoma brucei parasite and spread by tsetse flies. Occurring only in sub-Saharan Africa, HAT causes neurological damage and is fatal when left untreated. Surveillance and control programs have helped bring reported cases down sharply, from more than 25,000 in 2000 to 980 last year. WHO hopes the subspecies T. b. gambiense , which occurs in West and Central Africa and is responsible for more than 98% of cases, can be eliminated altogether by 2030. “I am sure [Togo's] efforts will inspire others,” Matshidiso Moeti, WHO regional director for Africa, said in a statement. 3.4 million —Square kilometers of sea floor changed by human activities, such as the construction of ports, communication cables, oil rigs, and wind farms, as of 2018, representing an estimated 1.5% of all coastal areas. ### Why it matters The modified area equals that of cities on land, and its marine ecosystems may have sustained damage ( Nature Sustainability ).

During July and August, UNESCO convened a global public online consultation, along with eleven regional and sub-regional virtual consultations, to discuss the first draft text of the Recommendation. It sought feedback that addressed various local concerns to achieve a truly inclusive and pluralist normative instrument. The consultations strengthened regional partnerships, particularly in Africa and Latin America and the Caribbean, and provided a platform to raise further social, economic, and cultural implications of AI globally: from a data breach to hate speech and harassment; from gender bias to AI accountability; from human rights to education and climate change. Addressing over 600 submissions and 50,000 suggestions – from policymakers, international organizations, civil society, media, private sector, academia, and the general public – twenty-four internationally renowned experts will re-examine ethical concerns in the emerging age of AI and draft appropriate revisions. All voices will be heard.

Artificial Intelligence has been credited as a major disruptive force leading us towards digital transformation. From helping in developing high-end robotics to weather predictions, stock market crash, drug discovery, to understanding customer data, better filling system, and so forth, AI is quite instrumental in the technological advancements we see today. Though it is a common thought that AI will rob us of our employment opportunities and jobs, in contrast, it has been observed that AI has helped us by augmenting our capabilities and assisting us in numerous activities, including the trivial ones. While we are all quite aware of how AI is shaping industries generally functions, there are some uncommon applications of AI too. Let us discuss some of them.

The "AI in Government - Forecasts from 2020 to 2025" report has been added to ResearchAndMarkets.com's offering. The Artificial Intelligence (AI) in government market was valued at US$4.904 billion in 2019. In recent years, government in different countries are taking a keen interest in artificial intelligence (AI) technology. They are increasingly investing in artificial intelligence (AI), spending budget, and time on pilot programs for various AI applications while discussing with people in the filed on the future implications of this technology for various public projects. The growing volume of big data is the major factor that is increasing the adoption of artificial intelligence (AI) technology across the government sector as it reduces the cost of storing and processing that data.

In massive multiple-input multiple-output (MIMO) systems, the large number of antennas would bring a great challenge for the acquisition of the accurate channel state information, especially in the frequency division duplex mode. To overcome the bottleneck of the limited number of radio links in hybrid beamforming, we utilize the neural networks (NNs) to capture the inherent connection between the uplink and downlink channel data sets and extrapolate the downlink channels from a subset of the uplink channel state information. We study the antenna subset selection problem in order to achieve the best channel extrapolation and decrease the data size of NNs. The probabilistic sampling theory is utilized to approximate the discrete antenna selection as a continuous and differentiable function, which makes the back propagation of the deep learning feasible. Then, we design the proper off-line training strategy to optimize both the antenna selection pattern and the extrapolation NNs. Finally, numerical results are presented to verify the effectiveness of our proposed massive MIMO channel extrapolation algorithm.

In this work we consider active local learning: given a query point $x$, and active access to an unlabeled training set $S$, output the prediction $h(x)$ of a near-optimal $h \in H$ using significantly fewer labels than would be needed to actually learn $h$ fully. In particular, the number of label queries should be independent of the complexity of $H$, and the function $h$ should be well-defined, independent of $x$. This immediately also implies an algorithm for distance estimation: estimating the value $opt(H)$ from many fewer labels than needed to actually learn a near-optimal $h \in H$, by running local learning on a few random query points and computing the average error. For the hypothesis class consisting of functions supported on the interval $[0,1]$ with Lipschitz constant bounded by $L$, we present an algorithm that makes $O(({1 / \epsilon^6}) \log(1/\epsilon))$ label queries from an unlabeled pool of $O(({L / \epsilon^4})\log(1/\epsilon))$ samples. It estimates the distance to the best hypothesis in the class to an additive error of $\epsilon$ for an arbitrary underlying distribution. We further generalize our algorithm to more than one dimensions. We emphasize that the number of labels used is independent of the complexity of the hypothesis class which depends on $L$. Furthermore, we give an algorithm to locally estimate the values of a near-optimal function at a few query points of interest with number of labels independent of $L$. We also consider the related problem of approximating the minimum error that can be achieved by the Nadaraya-Watson estimator under a linear diagonal transformation with eigenvalues coming from a small range. For a $d$-dimensional pointset of size $N$, our algorithm achieves an additive approximation of $\epsilon$, makes $\tilde{O}({d}/{\epsilon^2})$ queries and runs in $\tilde{O}({d^2}/{\epsilon^{d+4}}+{dN}/{\epsilon^2})$ time.

In this post, I interview my former UMBC lab mate, Dr. Huguens Jean, who was just hired to work at Google's Video AI Group as an artificial intelligence researcher. Huguens shares his inspirational story, starting from Port-au-Prince, Haiti where he was born and raised, to his schooling at UMBC, and now to his latest position at Google. He also shares details on his humanitarian efforts where he's successfully applied computer vision and deep learning to rural Rwanda to help count footfall traffic. The data him and his team gathered through footfall traffic analysis was used to help the non-profit organization, Bridges to Prosperity, to construct infrastructure such as bridges and roads, to better connect Rwanda villages. Let's give a warm welcome to Dr. Huguens Jean as he shares his story. Thank you for doing this interview. It's such a wonderful pleasure to have you here on the PyImageSearch blog.

Concerned that the majority of air quality issues affected those in low-income countries, the team developed a solution called AirQo, an initiative that combines human ingenuity, AI models, and boxes packed with air monitoring technology to predict pollution patterns in Kampala. Air sensors on buildings and moto-taxis collect swathes of pollution data, and cloud-based AI software swiftly analyses it to make air quality forecasts. These predictions are passed to government agencies, who can work to improve air quality and reduce the risk of exposure within local communities. The research team hopes that one day this technology will reduce pollution on streets across the continent so that this generation and the next will know what it means to breathe fresh air.

Many interactive data systems combine visual representations of data with embedded algorithmic support for automation and data exploration. To effectively support transparent and explainable data systems, it is important for researchers and designers to know how users understand the system. We discuss the evaluation of users' mental models of system logic. Mental models are challenging to capture and analyze. While common evaluation methods aim to approximate the user's final mental model after a period of system usage, user understanding continuously evolves as users interact with a system over time. In this paper, we review many common mental model measurement techniques, discuss tradeoffs, and recommend methods for deeper, more meaningful evaluation of mental models when using interactive data analysis and visualization systems. We present guidelines for evaluating mental models over time that reveal the evolution of specific model updates and how they may map to the particular use of interface features and data queries. By asking users to describe what they know and how they know it, researchers can collect structured, time-ordered insight into a user's conceptualization process while also helping guide users to their own discoveries.