Housed administratively in the College of IST, the Center will bring together researchers from across the University to develop new AI technologies and understand their social and ethical implications. The Penn State Center for Socially Responsible Artificial Intelligence promotes the thoughtful development and application of AI and studies its impact on all areas of human endeavor. In addition to supporting research focused explicitly on AI for social good and mitigating threats from its misuse, through this center, Penn State will encourage that all AI research and development activities consider social and ethical implications as well as intended and possible unintended consequences. "Given the rapid expansion and progression of interdisciplinary research and the wide-ranging impact of artificial intelligence on society, this center will engage and enable Penn State scholars and educators to work together and use AI to address the grand challenges of our time," said Andrew Sears, dean of the College of Information Sciences and Technology (IST), who led the founding of the center. The endeavor will bring together researchers from diverse disciplines across the University, enabling multidisciplinary research and educational programs that will shape the future of AI.

Guided by the goal of obtaining an optimization algorithm that is both fast and yielding good generalization, we study the descent direction maximizing the decrease in generalization error or the probability of not increasing generalization error. The surprising result is that from both the Bayesian and frequentist perspectives this can yield the natural gradient direction. Although that direction can be very expensive to compute we develop an efficient, general, online approximation to the natural gradient descent which is suited to large scale problems. We report experimental results showing much faster convergence in computation time and in number of iterations with TONGA (Topmoumoute Online natural Gradient Algorithm) than with stochastic gradient descent, even on very large datasets. Papers published at the Neural Information Processing Systems Conference.

Adaptation to other initially unknown agents often requires computing an effective counter-strategy. In the Bayesian paradigm, one must find a good counter-strategy to the inferred posterior of the other agents' behavior. In the experts paradigm, one may want to choose experts that are good counter-strategies to the other agents' expected behavior. In this paper we introduce a technique for computing robust counter-strategies for adaptation in multiagent scenarios under a variety of paradigms. The strategies can take advantage of a suspected tendency in the decisions of the other agents, while bounding the worst-case performance when the tendency is not observed.

You can find the files for this project at my GitHub and the slides here. The final project is accessible here (interactive web app).] I recently designed a new approach to automatic pricing for Airbnb listings using the Inside Airbnb dataset. I used linear regression to establish a base price and time series analysis to forecast price fluctuations due to the date. I used unsupervised learning to build a recommender system so hosts could compare their listing to other similar popular listings.

The new Pavilion invites AI hardware innovators to exhibit at DAC in a turnkey solution package SAN FRANCISCO, CA. – February 13, 2020 –The Design Automation Conference (DAC), the premier conference devoted to the design and automation of electronic circuits and systems, will this year showcase a dedicated Pavilion centered on the artificial intelligence (AI) hardware ecosystem. AI hardware is driving the largest wave of chip-design activity in decades. Understanding and harnessing the enormous computational and application potential of AI is fertile ground for new ideas and startup providers. Converting these ideas into working hardware circuits and systems is the core value of design automation, and the major technical focus of 57th DAC. The 57th DAC will be held at Moscone West Center in San Francisco, CA, from July 19-23, 2020.

A machine learning model's performance is only as good as the quality of the data set on which it's trained, and in the domain of self-driving vehicles, it's critical this performance isn't adversely impacted by errors. A troubling report from computer vision startup Roboflow alleges that exactly this scenario occurred -- according to founder Brad Dwyer, crucial bits of data were omitted from a corpus used to train self-driving car models. Dwyer writes that Udacity Dataset 2, which contains 15,000 images captured while driving in Mountain View and neighboring cities during daylight, has omissions. Thousands of unlabeled vehicles, hundreds of unlabeled pedestrians, and dozens of unlabeled cyclists are present in roughly 5,000 of the samples, or 33% (217 lack any annotations at all but actually contain cars, trucks, street lights, or pedestrians). Worse are the instances of phantom annotations and duplicated bounding boxes (where "bounding box" refers to objects of interest), in addition to "drastically" oversized bounding boxes.

In 2020, we will see US governments shift the conversation from who implements AI fastest to how we can implement most responsibly. While China is already using AI to measure students' brain waves with IoT sensors during class to help teachers provide more customizable content to achieve better retention and results, it's likely that the U.S. government will focus heavily in the coming year on privacy regulations to ensure AI use cases like this are fully vetted before being allowed. Federal regulations on privacy when it comes to the use of AI will take center stage in 2020. We've already seen the beginnings of this with two instances of the U.S. government taking action to prevent AI overstepping in states California and Massachusetts. This past May, the San Francisco Board of Supervisors banned the use of facial recognition technology by police and all other municipal agencies under the Stop Secret Surveillance Ordinance.

We consider problems for which one has incomplete binary matrices that evolve with time (e.g., the votes of legislators on particular legislation, with each year characterized by a different such matrix). An objective of such analysis is to infer structure and inter-relationships underlying the matrices, here defined by latent features associated with each axis of the matrix. In addition, it is assumed that documents are available for the entities associated with at least one of the matrix axes. By jointly analyzing the matrices and documents, one may be used to inform the other within the analysis, and the model offers the opportunity to predict matrix values (e.g., votes) based only on an associated document (e.g., legislation). The research presented here merges two areas of machine-learning that have previously been investigated separately: incomplete-matrix analysis and topic modeling.

Automation fears distract from the real problem: too few blue-collar workers. This article is part of an MIT SMR initiative exploring how technology is reshaping the practice of management. Following the Great Recession, anxiety intensified over the prospect of automation causing permanent, widespread unemployment. Feeding on public alarm, a large number of studies assessed the likely impact of future automation on jobs. Although some touted the potential for job creation, others predicted catastrophic job loss. Today, after more than a decade of continuous U.S. economic expansion, the fear of automation remains entrenched in the country's psyche, dominating public discussions and political debates.

Artificial intelligence has accurately predicted the possibility of heart attack or stroke in a world's first. A study led by Barts Health NHS Trust and University College London used AI to analyse cardiac scans of more than 1,000 patients. Researchers said it's the first time blood flow scans, which reveal problems with the heart, have been read by a computer. The technology was more accurate at predicting major cardiovascular events within a 19-month follow-up than a doctor using traditional means. Researchers said it could be used by medical teams to recommend treatments.