Linear relaxation based perturbation analysis for neural networks, which aims to compute tight linear bounds of output neurons given a certain amount of input perturbation, has become a core component in robustness verification and certified defense. However, the majority of linear relaxation based methods only consider feed-forward ReLU networks. While several works extended them to relatively complicated networks, they often need tedious manual derivations and implementation which are arduous and error-prone. Their limited flexibility makes it difficult to handle more complicated tasks. In this paper, we take a significant leap by developing an automatic perturbation analysis algorithm to enable perturbation analysis on any neural network structure, and its computation can be done automatically in a similar manner as the back-propagation algorithm for gradient computation. The main idea is to express a network as a computational graph and then generalize linear relaxation algorithms such as CROWN as a graph algorithm. Our algorithm itself is differentiable and integrated with PyTorch, which allows to optimize network parameters to reshape bounds into desired specifications, enabling automatic robustness verification and certified defense. In particular, we demonstrate a few tasks that are not easily achievable without an automatic framework. We first perform certified robust training and robustness verification for complex natural language models which could be challenging with manual derivation and implementation. We further show that our algorithm can be used for tasks beyond certified defense - we create a neural network with a provably flat optimization landscape and study its generalization capability, and we show that this network can preserve accuracy better after aggressive weight quantization. Code is available at https://github.com/KaidiXu/auto_LiRPA.

We propose a method for estimation in high-dimensional linear models with nominal categorical data. Our estimator, called SCOPE, fuses levels together by making their corresponding coefficients exactly equal. This is achieved using the minimax concave penalty on differences between the order statistics of the coefficients for a categorical variable, thereby clustering the coefficients. We provide an algorithm for exact and efficient computation of the global minimum of the resulting nonconvex objective in the case with a single variable with potentially many levels, and use this within a block coordinate descent procedure in the multivariate case. We show that an oracle least squares solution that exploits the unknown level fusions is a limit point of the coordinate descent with high probability, provided the true levels have a certain minimum separation; these conditions are known to be minimal in the univariate case. We demonstrate the favourable performance of SCOPE across a range of real and simulated datasets. An R package CatReg implementing SCOPE for linear models and also a version for logistic regression is available on CRAN.

Self-driving cars are one of the high-risk artificial intelligence applications the European Union wants to regulate. The European Commission today unveiled its plan to strictly regulate artificial intelligence (AI), distinguishing itself from more freewheeling approaches to the technology in the United States and China. The commission will draft new laws--including a ban on "black box" AI systems that humans can't interpret--to govern high-risk uses of the technology, such as in medical devices and self-driving cars. Although the regulations would be broader and stricter than any previous EU rules, European Commission President Ursula von der Leyen said at a press conference today announcing the plan that the goal is to promote "trust, not fear." The plan also includes measures to update the European Union's 2018 AI strategy and pump billions into R&D over the next decade. The proposals are not final: Over the next 12 weeks, experts, lobby groups, and the public can weigh in on the plan before the work of drafting concrete laws begins in earnest.

SAN FRANCISCO – As companies quickly adopt machine learning systems, cybercriminals are close behind scheming to compromise them. That worries legal experts who say a lack of laws swing open the door for bad guys to attack systems. During a panel session at RSA Conference 2020 this week, Cristin Goodwin, the assistant general counsel with Microsoft, said the number of machine learning related U.S. court cases is a mere 52. She noted most were related to patents, workplace discrimination and even gerrymandering. Few court cases addressed actual cyberattacks on machine learning systems – demonstrating a dangerous dearth in legal precedent around the technology.

The U.S. Energy Department earlier this month appointed former 3M Co. artificial-intelligence leader Cheryl Ingstad as the first director of its Artificial Intelligence and Technology Office, where she will oversee the DOE's AI activities. The mission of the AITO, which was formed in September 2019, is to coordinate the department's artificial-intelligence activities, which includes scaling AI projects across the DOE, sharing best practices and reducing duplicate projects. The office also is charged with facilitating partnerships...

News Top Stories'Staying True to the Medicaid Promise' Top Stories SUNY Polytechnic Institute artificial intelligence project Top Stories Little Falls Woman Pleads Guilty To Bank Fraud Utica Police ID Second Person In Murder–Attempted Suicide Traffic violations in NYS discovery law Video Whitesboro Flood Meeting Thursday Night Video Weather Bank of Utica EYEnet Sports Top Stories Yankees' Stanton Injured, Unlikely for Opening Day Video Comets Beat Americans in Shootout, Rafferty Breaks Record Video Oneida Girls Basketball Dominates Against Lowville in Quarterfinals – Highlights Video Colgate Women's and Men's Basketball Fall to Bucknell Video Community Remarkable Women Full Interviews Daily Pledge of Allegiance Contests Rain – Beatles Tribute Contest About Us Do Not Sell My Personal Information Rep. Brindisi announces re-election campaign US women's pursuit takes gold at track cycling championships Ohio State star Chase Young follows in Nick Bosa's footsteps

How Artificial Intelligence and Automation may reshape supply chain roles as we know them? Technological advancements throughout time have caused periods of huge change.

The Defense Department has officially adopted a set of principles to ensure ethical artificial intelligence adoption, but much work is needed on the implementation front, senior DOD tech officials told reporters Feb. 24. The five principles [see sidebar], which are based on the recommendations of the Defense Innovation Board's 15-month study on the matter, represent a first step and generalized intentions around AI use and adoption including being responsible, equitable, traceable, reliable, and governable. DOD released the principles during a news briefing Feb. 24. Those AI ethical guidelines will likely be woven into a little bit of everything, like cyber, from data collection to testing, DOD CIO Dana Deasy told reporters. "We need to be very thoughtful about where that data is coming from, what was the genesis of that data, how was that data previously being used and you can end up in a state of [unintentional] bias and therefore create an algorithmic outcome that is different than what you're actually intending," Deasy said.

Clearview AI, a start-up with a database of more than three billion photographs from Facebook, YouTube and Twitter, has been hacked. The attack allowed hackers to gain access to its client list but it said its servers had not been breached. Most of its clients are US law enforcement agencies who use its facial-recognition software to identify suspects. Its use of images scraped from the internet has raised privacy concerns. The company told BBC News: "Security is Clearview's top priority. "Unfortunately, data breaches are part of life in the 21st Century.

Artificial intelligence (AI) is changing the way we live and work, and it has the potential to transform virtually every sector of the global economy – from healthcare and education to transport and energy systems. It holds tremendous promise for economic productivity, scientific advancements and sustainable development. Yet there are also anxieties around where the technology may lead us, including concerns that it may codify existing biases and discriminatory practices, or displace workers by accelerating automation. For governments, this presents a pressing question: how can we harness the potential of AI to improve well-being for all, while also mitigating the risks that it poses? The OECD AI Principles – the first intergovernmental standard on AI – provide a way forward.