We propose a reward function estimation framework for inverse reinforcement learning with deep energy-based policies. We name our method PQR, as it sequentially estimates the Policy, the $Q$-function, and the Reward function by deep learning. PQR does not assume that the reward solely depends on the state, instead it allows for a dependency on the choice of action. Moreover, PQR allows for stochastic state transitions. To accomplish this, we assume the existence of one anchor action whose reward is known, typically the action of doing nothing, yielding no reward. We present both estimators and algorithms for the PQR method. When the environment transition is known, we prove that the PQR reward estimator uniquely recovers the true reward. With unknown transitions, we bound the estimation error of PQR. Finally, the performance of PQR is demonstrated by synthetic and real-world datasets.

As roles and tasks shift in tandem with the expansion of new technologies, and the division of work between human and machine is redrawn, it is of critical importance to monitor how those changes will impact the evolution of economic gender gaps. Artificial Intelligence (AI) is a prominent driver of change within the transformations brought about by the Fourth Industrial Revolution (4IR), and can serve as key marker of the trajectory of innovation across industries.19 In partnership with the LinkedIn Economic Graph Team, the World Economic Forum aims to provide fresh evidence of the emerging contours of gender parity in the new world of work through near-term labour market information. The increasing expansion of AI is creating the demand for a range of new skills, among them neural networks, deep learning, machine learning, and "tools" such as Weka and Scikit-Learn. AI skills are among the fastest-growing specializations among professionals represented on the LinkedIn platform.

—Rebecca Bayeck and Azure Stewart “Artificial Intelligence and Archives” was the inaugural webinar of the series on Emerging Technologies, Big Data & Archives, organized by CLIR postdocs Rebecca Y. Bayeck of the Schomburg Center for Research in Black Culture and Azure Stewart of New York University. With the emergence of new technologies and big data, the processing and preservation of data has changed and will continue to change. As in other domains (e.g., health, video games), artificial intelligence (AI) is increasingly reshaping the way we process, interact with, and think about archives. Consequently, in the age of big data, archives are not just “a collection of historical records relating to a place, organization, or family” (Cambridge Dictionary Online). Today, archives also include all types of digital data—including social media data—and algorithms. Archivists are therefore called on to preserve and process data as they are being created, which requires understanding AI languages, processes, and practices for the creation and protection of data/records now for the future. In this webinar, our speaker Dr. Anthea Seles, from the International Council on Archives (ICA), discussed AI in archival spaces: its uses, application, and the role archivists should play to become critical voices in AI discussions. Two hours were not enough to address all the questions raised by the 280 attendees. As a follow up to the webinar, we have thematically organized and addressed the unanswered questions and present them here. Artificial Intelligence in Archives How much has AI penetrated archives in the developing world? I would say [this has been] limited, if at all. I think the main issue is that these technologies are being applied in the assessment of development initiatives like Sustainable Development Goals (SDGs). Increasingly there are many projects focusing on artificial intelligence and human rights, for example the University of Essex Human Rights, Big Data and Technology Project, and it is becoming a concern for organisations like Amnesty International. Who already has the best AI for archives today, according to ICA regulation, that we can adopt? There is no commercial provider that works specifically on archival questions. I think you can use off-the-shelf eDiscovery software, but you need to have a basic understanding of what the technology is doing in order to measure your precision and recall.  Artificial Intelligence Tools Will governments and big corporations use artificial intelligence as a tool to centralize information in future? Potentially. I think there is some thinking about this coming out of the records management community, but I still believe it is about balancing the strengths of the tool with the continuing need for human intervention. The question is, when will the human be needed? And what can the tool be trusted to do with minimum supervision? How do we ensure a continuous feedback loop to identify records of long-term value as information creation changes?  What tools were you using for the file analysis and visualization in this presentation? The screen shots are only example photos, they are not from any of the tools we used. We looked at several eDiscovery tools with different algorithms (e.g., Latent Semantic Indexing, Latent Dirichlet Allocation). These are bog standard machine learning applications that have been around for a while, and we chose to go down that road to see what we could get in off-the-shelf commercial software packages. So, is there a way to write a script to avoid metadata corruption and alteration? There are tools now you can use that will preserve the integrity of the metadata when you move material from one system or file to another. I think for historical metadata alteration/corruption it is a question of how we explain this to users and how this might affect different access methods like visualisation.  Will the International Council on Archives provide training on artificial intelligence and machine learning? Not yet, but I’m open to suggestions. [We are] currently speaking with different stakeholders and maybe we can hold a hackathon at the Abu Dhabi Congress.  Access to Archives Will the course Managing Digital Archives be accessible online? The managing digital archives course is organized by the ICA and will be accessible online in fall 2020. Please check the ICA website or social media channels (Twitter and Facebook) for more information. What are some of the practices in the UK National Archives and government on managing structured data as records? How does the UK identify, capture, manage, and apply retention and disposition to data (both transactional applications and analytical ones)? There are no published policies on identification of datasets that I can see and would suggest you contact either the record copying or the UK government web archive records unit to see if anything more substantive has been developed. What is your suggestion for keeping physical records for posterity and authentication? Records should always be maintained in the format in which they are created. The belief in scanning paper records and destroying them in order to save space and save on storage costs is a false economy. The level at which you should be scanning that material and the amount of metadata that should be captured to maintain it over time is very high. Also, you need to take into account computer storage costs, and whether you can afford the costs of digital preservation software, which all begins to add up. One must also take into account the active management of these authentic digital surrogates by digital preservation specialists. Furthermore, if you have a paper management problem and you don’t take that into account when you move into the digital environment you are then transferring an analog integrity issue into a digital integrity/authenticity issue. Digital will not solve integrity issues; in my opinion it will magnify them. Artificial Intelligence and Society In Brazil, we are concerned with the problem of the spread and political use of misinformation (fake news). How can archivists with algorithm training provide reliable research insights to fight against this historical problem? At this point, I couldn’t honestly provide you with an answer but Read More

One-way car-sharing systems are transportation systems that allow customers to rent cars at stations scattered around the city, use them for a short journey, and return them at any station. The maximum customers' satisfaction problem concerns the task of assigning the cars, initially located at given stations, to maximize the number of satisfied customers. We consider the problem with two stations where each customer has exactly two demands in opposite directions between both stations, and a customer is satisfied only if both their demands are fulfilled. For solving this problem, we propose mixed-integer programming (MIP) models and matheuristics based on local search. We created a benchmark of instances used to test the exact and heuristic approaches. Additionally, we proposed a preprocessing procedure to reduce the size of the instance. Our MIP models can solve to optimality 85% of the proposed instances with 1000 customers in 10 minutes, with an average gap smaller than 0.1% for all these instances. For larger instances (2500 and 5000 customers), except for some particular cases, they presented an average gap smaller than 0.8%. Also, our local-based matheuristics presented small average gaps which are better than the MIP models in some larger instances.

The COVID-19 pandemic has been a major challenge to humanity, with 12.7 million confirmed cases as of July 13th, 2020 [1]. In previous work, we described how Artificial Intelligence can be used to tackle the pandemic with applications at the molecular, clinical, and societal scales [2]. In the present follow-up article, we review these three research directions, and assess the level of maturity and feasibility of the approaches used, as well as their potential for operationalization. We also summarize some commonly encountered risks and practical pitfalls, as well as guidelines and best practices for formulating and deploying AI applications at different scales.

Previous work in the social sciences, psychology and linguistics has show that liars have some control over the content of their stories, however their underlying state of mind may "leak out" through the way that they tell them. To the best of our knowledge, no previous systematic effort exists in order to describe and model deception language for Brazilian Portuguese. To fill this important gap, we carry out an initial empirical linguistic study on false statements in Brazilian news. We methodically analyze linguistic features using the Fake.Br corpus, which includes both fake and true news. The results show that they present substantial lexical, syntactic and semantic variations, as well as punctuation and emotion distinctions.

The roleTired of data science roles that end up being mostly data engineering? Or spending years on edge cases for a single application?Join our 100% remote, full-time, Machine Learning team, and work on exciting ML solutions every day.Take responsibility for the entire ML project cycle: From re...

In a recruitment industry, selecting a best CV from a particular job post within a pile of thousand CV's is quite challenging. Finding a perfect candidate for an organization who can be fit to work within organizational culture is a difficult task. In order to help the recruiters to fill these gaps we leverage the help of AI. We propose a methodology to solve these problems by matching the skill graph generated from CV and Job Post. In this report our approach is to perform the business understanding in order to justify why such problems arise and how we intend to solve these problems using natural language processing and machine learning techniques. We limit our project only to solve the problem in the domain of the computer science industry.

Behavioral cloning is an imitation learning technique that teaches an agent how to behave through expert demonstrations. Recent approaches use self-supervision of fully-observable unlabeled snapshots of the states to decode state-pairs into actions. However, the iterative learning scheme from these techniques are prone to getting stuck into bad local minima. We address these limitations incorporating a two-phase model into the original framework, which learns from unlabeled observations via exploration, substantially improving traditional behavioral cloning by exploiting (i) a sampling mechanism to prevent bad local minima, (ii) a sampling mechanism to improve exploration, and (iii) self-attention modules to capture global features. The resulting technique outperforms the previous state-of-the-art in four different environments by a large margin.

In the wildlife reserves of East Africa, elephants, rhinos, gorillas, and other large mammals are hunted by poachers. All that stands between these animals and harm's way are small teams of park rangers and conservationists. The danger is very real for these species on the brink: A staggering 35,000 African elephants are killed each year, putting them just a decade away from extinction, according to the non-profit RESOLVE. Technology is an increasingly critical tool for protecting elephants and other large animals, given their necessarily expansive habitats: A group of just 50 rangers in Kenya, for example, covers a reserve of 3,000 square miles. Park rangers and conservationists have used motion-activated camera traps to catch poachers in action, but the animals are tragically already lost by the time rangers can respond.