Reinforcement Learning Aids Fusion Control

Linking AI Methods to  Nuclear fusion energy. 

Exploring Reinforcement Learning to Control Nuclear Fusion Reactions

Carnegie Mellon University News

Aaron Aupperlee, September 8, 2022

Carnegie Mellon University (CMU) doctoral candidate Ian Char, the first CMU researcher to run an experiment on the DIII-D National Fusion Facility's tokamak machine, demonstrated that reinforcement learning algorithms can control the rotation of the machine's hydrogen plasma. Char developed two algorithms: one was trained using data from the tokamak on how the plasma reacts, while the other calculates the rate and direction at which to add hydrogen particles to affect the speed of the plasma's rotation. Said CMU's Jeff Schneider, "This work shows a path to using reinforcement learning to control other parts of the plasma state and ultimately achieve the temperatures and pressures long enough to have a power plant. That would mean limitless, clean energy for everyone."

The Simulated Kiss in VR

 New Experiences in VR 

VR Researchers Have Basically Figured Out How to Simulate the Feel of Kisses

A modified VR headset can create the sensation of touch, either on a user's lips or even inside their mouths.

By Andrew Liszewski  in Gizmodo

Without adding any hardware that actually makes contact with the wearer’s face, researchers from Carnegie Mellon University’s Future Interfaces Group have modified an off-the-shelf virtual reality headset so that it recreates the sensation of touch in and around a user’s mouth, finally fulfilling virtual reality’s inevitable one true purpose.

Aside from handheld controllers that occasionally vibrate, most consumer-ready virtual reality devices ignore senses like taste, smell, and touch, and instead focus on visuals and sounds. It’s enough to make virtual reality experiences far more compelling than they were decades ago, but not enough to truly fool the brain into thinking that what your eyes are seeing is possibly a real-life experience.  ... ' 

See also:  https://www.figlab.com/research/2022/mouth-haptics

Fingerprinting the IOT

Fingerprinting the IoT

Carnegie Mellon University College of Engineering News

Madison Brewer, February 9, 2022

Carnegie Mellon University (CMU) researchers strengthened the security of Internet of Things (IoT) devices by making them more resilient against exploitation through their development of radio-frequency fingerprinting (RFF). RFF can be used to identify specific IoT devices by detecting hardware variations that produce unique radio wave signatures. CMU's Jiachen Xu used power amplifiers to foil RFF exploits by changing the IoT signal's features, and a convolutional neural network classified incoming signals as safe or unsafe by assessing the RFF in the processed signal. The researchers also proved Bayesian neural networks could identify and classify RFF quickly and accurately, without requiring excessive computational power.

 Full article

CMU Cloud Lab Links Automation to Labor

 The linking of labor intensive needs is particularly interesting to testallternatives. 

Carnegie Mellon's Cloud Lab to Automate Labor-Intensive Science Experiments

The Wall Street Journal, Sara Castellanos,  August 30, 2021

Carnegie Mellon University (CMU) and remote technology access provider Emerald Cloud Lab have partnered on a $40-million laboratory to automate experiments in robotics and artificial intelligence. CMU's Rebecca Doerge said the "Cloud Lab" will be the first-ever university-owned cloud lab, in which scientists may access software to design experiments that are conducted remotely by about 200 different kinds of robots. Emerald Cloud Lab's Brian Frezza said researchers can design experiments from anywhere with the company's proprietary software, which instructs machines via the cloud. CMU researchers are developing artificial intelligence systems that can automatically suggest ways to calibrate the equipment to carry out tests optimally. .... 

Training Autonomous Drones

More advances in the use of data to train autonomous drones in many contexts.   Likely to see many more autonomous drone applications.   Moving the sensors to the data.

CMU Researchers Train Autonomous Drones Using Cross-Modal Simulated Data
Carnegie Mellon University
Virginia Alvino Young
August 25, 2020

Researchers at Carnegie Mellon University (CMU) developed a two-step approach to teaching autonomous drones perception and action, providing a safe way to deploy drones trained entirely on simulated data into real-world course navigation. In the first step, the researchers used a photorealistic simulator to train the drone on image perception by creating an environment including the drone, a soccer field, and elevated red square gates positioned randomly to create a track. Thousands of randomly generated drone and gate configurations were used to create a large dataset employed in the second step to teach the drone perception of positions and orientations in space. Said CMU's Rogerio Bonatti, "The robot is not learning to recreate going through any specific track. Rather, by strategically directing the simulated drone, it's learning all of the elements and types of movements to race autonomously." .... ' 

Bot Activity During Coronavirus

Don't know what to fully make of this.    How accurate is the machine learning of the model working to identify bots?.  Looking for the Carnegie piece supporting this to get an idea.  Here is one CMU article which covers the research.

Researchers: Nearly Half Of Accounts Tweeting About Coronavirus Are Likely Bots   By Bobby Allyn

Computer scientists at Carnegie Mellon University have determined that nearly half of all Twitter accounts spreading messages about the COVID-19 pandemic are likely bots. The team analyzed more than 200 million tweets discussing the virus since January, and found about 45% were sent by accounts that behave more like computerized bots than humans. In addition, the researchers identified more than 100 false narratives about the novel coronavirus that bot-controlled accounts are spreading on the platform. The researchers used a bot-hunter tool to flag accounts that post messages more often than is humanly possible, or which claim to be in multiple countries within a period of a few hours. Said Carnegie Mellon researcher Kathleen Carley, "We're seeing up to two times as much bot activity as we'd predicted based on previous natural disasters, crises, and elections."   ... '

How to Learn without Supervision

Ultimately technical, but note the similarity to the alternate ways we learn ... by rote memorization, and the continuous repetition in a contextual background.  Confirmed or not by the context.  Useful for different things.  The latter is more common, but also for different kinds of tasks.

Unsupervised Meta-Learning: Learning to Learn without Supervision
By Benjamin Eysenbach and Abhishek Gupta    May 1, 2020
This post is cross-listed on the CMU ML blog.

The history of machine learning has largely been a story of increasing abstraction. In the dawn of ML, researchers spent considerable effort engineering features. As deep learning gained popularity, researchers then shifted towards tuning the update rules and learning rates for their optimizers. Recent research in meta-learning has climbed one level of abstraction higher: many researchers now spend their days manually constructing task distributions, from which they can automatically learn good optimizers. What might be the next rung on this ladder? In this post we introduce theory and algorithms for unsupervised meta-learning, where machine learning algorithms themselves propose their own task distributions. Unsupervised meta-learning further reduces the amount of human supervision required to solve tasks, potentially inserting a new rung on this ladder of abstraction.

We start by discussing how machine learning algorithms use human supervision to find patterns and extract knowledge from observed data. The most common machine learning setting is regression, where a human provides labels Y for a set of examples X. The aim is to return a predictor that correctly assigns labels to novel examples. Another common machine learning problem setting is reinforcement learning (RL), where an agent takes actions in an environment. In RL, humans indicate the desired behavior through a reward function that the agent seeks to maximize. To draw a crude analogy to regression, the environment dynamics are the examples X, and the reward function gives the labels Y. Algorithms for regression and RL employ many tools, including tabular methods (e.g., value iteration), linear methods (e.g., linear regression) kernel-methods (e.g., RBF-SVMs), and deep neural networks. Broadly, we call these algorithms learning procedures: processes that take as input a dataset (examples with labels, or transitions with rewards) and output a function that performs well (achieves high accuracy or large reward) on the dataset. ... "

Making Systems Aware of Context

Awareness, also known as context, is often an important aspect of intelligence.

Two Systems Allow Smart Devices to Have Environmental Awareness
in Electronics360    By Siobhan Treacy

Two teams of Carnegie Mellon University (CMU) researchers have developed two new systems that give smart devices environmental awareness by analyzing either sound or vibrations to register their surroundings. The Ubicoustics system uses the microphones in smart devices to register and classify environmental sounds, having been trained on sound-effect archives used by the entertainment industry. Says CMU's Gierad Laput, "We can transform and project [these sound libraries] into hundreds of different variations, creating volumes of data perfect for training deep learning models." The other system, Vibrosight, classifies sounds by detecting vibrations via an integrated sensor, laser, and motorized steerable mirror. Ubicoustics and Vibrosight were both presented this week at the ACM User Interface Software and Technology Symposium (UIST2018) in Berlin. .... " 

Computer Controlled Knitting

Like the unusual process method, makes us think of the possibility of others.

Software Automatically Generates Knitting Instructions for 3D Shapes
Carnegie Mellon University
Byron Spice

Carnegie Mellon University (CMU) researchers have developed a system that can translate various three-dimensional (3D) shapes into stitch-by-stitch instructions executed by a computer-controlled knitting machine, and have used it to produce plush toys and garments. CMU professor James McCann envisions this milestone perhaps leading to on-demand machine knitting. His team's algorithm takes the knitting machines' limitations into account, generating instructions for patterns that work within these constraints and lower the risk of yarn breaks or jams. McCann notes additional work is needed to expand the system's capabilities beyond producing only smooth knitted cloth, which lacks the patterned stitching that can make knitted garments distinctive. "The software...needs a little push," McCann says, "and software can improve rapidly because we can iterate so much faster." The work will be presented in August at the ACM Conference on Computer Graphics and Interactive Techniques (SIGGRAPH 2018) in Vancouver, Canada. ... " 

Never-Ending Learning Continues at CMU

Was reminded of this effort.  During our early work with AI, based on very large and messy rule bases, we tried to introduce learning capabilities.   I note this has been going on at CMU since 2010.  How might this connect with other kinds of applications like CYC?   Ultimately such efforts will build the foundation of broad AI.  Here an update, examining further and will now stay in touch.

Though technical, some useful discussion of the underlying motivations.

Here the site for the project, which includes a simple statement of purpose.

A paper of depth with statements of progress:  Never-Ending Learning

By T. Mitchell, W. Cohen, E. Hruschka, P. Talukdar, B. Yang, J. Betteridge, A. Carlson, B. Dalvi, M. Gardner, B. Kisiel, J. Krishnamurthy, N. Lao, K. Mazaitis, T. Mohamed, N. Nakashole, E. Platanios, A. Ritter, M. Samadi, B. Settles, R. Wang, D. Wijaya, A. Gupta, X. Chen, A. Saparov, M. Greaves, J. Welling 

Communications of the ACM, Vol. 61 No. 5, Pages 103-115
10.1145/3191513

Whereas people learn many different types of knowledge from diverse experiences over many years, and become better learners over time, most current machine learning systems are much more narrow, learning just a single function or data model based on statistical analysis of a single data set. We suggest that people learn better than computers precisely because of this difference, and we suggest a key direction for machine learning research is to develop software architectures that enable intelligent agents to also learn many types of knowledge, continuously over many years, and to become better learners over time. In this paper we define more precisely this never-ending learning paradigm for machine learning, and we present one case study: the Never-Ending Language Learner (NELL), which achieves a number of the desired properties of a never-ending learner. NELL has been learning to read the Web 24hrs/day since January 2010, and so far has acquired a knowledge base with 120mn diverse, confidence-weighted beliefs (e.g., servedWith(tea,biscuits)), while learning thousands of interrelated functions that continually improve its reading competence over time. NELL has also learned to reason over its knowledge base to infer new beliefs it has not yet read from those it has, and NELL is inventing new relational predicates to extend the ontology it uses to represent beliefs. We describe the design of NELL, experimental results illustrating its behavior, and discuss both its successes and shortcomings as a case study in never-ending learning. NELL can be tracked online at http://rtw.ml.cmu.edu, and followed on Twitter at @CMUNELL.  ... "

Giant Touchpads

Talked to office furniture manufactures about some of these methods.

A coat of conductive paint can transform regular walls into giant touchpads   In Digital Trends By Luke Dormehl

The late Mark Weiser, the former chief scientist at Xerox PARC, once stated, “The most profound technologies are those that disappear. They weave themselves into the fabric of everyday life until they are indistinguishable from it.” Few tech projects summarize that vision of ubiquitous computing better than a new project carried out by researchers from Carnegie Mellon University and Disney Research. Through the use of conductive paint and some low-cost electronics, they have found a way to transform regular, boring internal walls into smart infrastructure able to detect human touch. .... "

Sony, CMU looking at Cooking Robots

The idea of further looking at task oriented interaction with recipes, techniques and recipes.  As a long time cook, and having worked in the Food industry.  Also involves close cooperation with people.  Been examining how assistants work with recipes.  Its a natural place to think about AI.  Lots of data, understood goals, sub-tasks and resources readily defined. 

Sony, Carnegie Mellon form partnership to research cooking robots
The skills needed for food preparation and delivery could be applied to a number of other industries and tasks, Sony and CMU said.      By Stephanie Condon for Between the Lines

Sony, Carnegie Mellon Form Partnership to Research Cooking Robots 

" ... Researchers at Carnegie Mellon University (CMU) and Sony are working together to study how artificial intelligence (AI) and robotics can be used to improve food preparation, cooking, and delivery. The researchers want to teach machines to handle fragile and irregularly shaped materials, skills that could be applied to other household or small business tasks. The project, which be run primarily out of CMU's School of Computer Science, has the potential to make the vast possibilities of AI and robotics more familiar and accessible to the general public, according to project lead Hiroaki Kitano, president and CEO of Sony Computer Science Laboratories. ... " 

Towards Crowd Sourcing Conversational Agents

Intriguing thought, not sure it actually is the same as what we did and called a 'Concierge' model, linking appropriately to other smart agents or people.   More like Facebook's attempt to include humans within the now defunct Facebook M.   Interesting experiment with the examples.  Once again using Amazon's  Mechanical Turk to crowd source.  And again our own experience was that the careful setup is key.

Crowd Workers, AI Make Conversational Agents Smarter 
Carnegie Mellon News,   By Byron Spice

Researchers at Carnegie Mellon University (CMU) have developed Evorus, a chatbot system that recruits crowd workers on demand from Amazon Mechanical Turk to answer questions from users, with the crowd workers voting on the best answer. Evorus also tracks the questions that have been asked and answered, and over time, it will begin to suggest these answers for subsequent questions. 

The researchers also developed a process by which the artificial intelligence (AI) can help to approve a message with less crowd worker involvement. During a five-month deployment, Evorus worked with 80 users and 181 conversations, and its automated responses to questions were chosen 12 percent of the time, crowd voting was reduced by almost 14 percent, and the cost of crowd work for each reply to a user's message dropped by 33 percent. The researchers will present Evorus in April at the ACM Conference on Human Factors in Computing Systems (CHI 2018) in Montreal, Canada. .... " 

Detecting Review and Follower Fraud

From CMU.  Tested against some very large set of Twitter data.   Uses a form of Graph Analytics.  Python code is available as Open Source.  Uses a form of  Faloutsos' NetProbe  (technical paper) for the analysis.

New algorithm detects online fraudsters: Method sees through camouflage to reveal fake followers, reviewers.

The method, called FRAUDAR, marks the latest escalation in the cat-and-mouse game played by online fraudsters and the social media platforms that try to out them. In particular, the new algorithm makes it possible to see through camouflage that fraudsters use to make themselves look legitimate, said Christos Faloutsos, professor of machine learning and computer science.

In real-world experiments using Twitter data for 41.7 million users and 1.47 billion followers, FRAUDAR fingered more than 4,000 accounts not previously identified as fraudulent, including many that used known follower-buying services such as TweepMe and TweeterGetter.

"We're not identifying anything criminal here, but these sorts of frauds can undermine people's faith in online reviews and behaviors," Faloutsos said. He noted most social media platforms try to flush out such fakery, and FRAUDAR's approach could be useful in keeping up with the latest practices of fraudsters.

The CMU algorithm is available as open-source (Python) code at http://www.andrew.cmu.edu/user/bhooi/camo.zip. A research paper describing the algorithm won the Best Paper Award last month at the Association for Computing Machinery's Conference on Knowledge Discovery and Data Mining (KDD2016) in San Francisco. ..... " 

Google and Carnegie form an IOT Campus

In Fast Company:  Seeking to figure out how things should communicate.  Good Goal.  Keep industry involved. Integrate cognitive intelligence.

" ... For the "Internet of Things" to thrive, all it needs is for all devices to get along—which is currently wishful thinking. Last week, however, Google announced a partnership with Carnegie-Mellon University, which is leading a collaboration of faculty from several other academic institutions on a project to jumpstart the Internet of Things revolution. Their plan: Build a universal platform that lets any device talk to any other device. And fittingly, that master-key solution will be open source.

The problem is that IoT software and devices are mostly proprietary, built by each company and working well within their own sandboxes, but they don't communicate well together. The joint project between CMU, Cornell, Stanford, Illinois at Urbana-Champaign, and Google wants to wipe away the private-industry middlemen that keep sensors in separate sandboxes by creating a new, open platform: GIoTTO.  .. " 

In the CMU News.

Communicating Robots

Note has some similarity to swarm robotics    But here the individuals are likely to be more specialized.    Collaborating,  But need for central management?   Here some examples from CMU.

In TechologyReview:
Making Robots Talk to Each , other  By Julia Sklar 
Robots that can work together and communicate are not only more efficient, they’re cheaper—since not every robot has to excel at every task. ... " 

kWantera: Intelligent Energy Analytics

Brought to my attention: kWantera   Impressive ideas.  With origins at CMU,  They write:

" ... kWantera is changing the way energy is bought, sold, and consumed with the goal of providing lower costs for buyers and higher profitability for sellers. The flexible yet complex nature of wholesale power markets gives almost anyone the opportunity, but no useful tools, to capture the daily savings and profit opportunities in these evolving markets. With kWantera, anyone can now reap the benefits of wholesale markets with accurate, consistent, and customized recommendations built on a data-driven understanding of the financial and physical market opportunities and risks in North America and the European Union energy markets. kWantera gives its customers the opportunity to make smarter daily buy, sell, and load management decisions through real-time predictive analytics. All of our recommendations are risk adjusted as well as time and location specific so our customers have the confidence to avoid unnecessary risks and make easy day-to-day energy decisions.  ... " 

Cognitive Agents in the Wild... the Jeopardy Experience

Excellent CSIG talk today: Slides here.   " ... Our speaker for this week is Eric Nyberg from Carnegie Mellon University, who will present " From Jeopardy! To Cognitive Agents: Effective Learning in the Wild". ... "  They have made considerable progress since we worked with them in the 90s.

Eric Nyberg's home page, with many more links.

Long Range Iris Scanning

We examined iris scanning for security applications.  You had to place your face on the reader, and it thus created a sanitary issue.  Also, some people could not be scanned because of face-eye structure. Now in the Atlantic:  " ...   Long-Range Iris Scanning Is Here ... An engineering professor at Carnegie Mellon says he’s invented technology that can identify someone from across the room with the precision of a fingerprint. ... "

How Enterprises Use AI

In CWorld:   Good overview of the expansion of the use of cognitive, AI and links to Machine learning.  Based on learnings at the recent AAAI meeting.   My experience and connections, have led to getting lots of queries about where to start.

" .... While A.I. is being used in smartphones and self-driving cars, it's also working its way into the enterprise to filter spam out of email, handle complicated scheduling or detecting fraud in big data deployments.

"I think IT probably needs artificial intelligence," said Stephen Smith, a professor who specializes in robotics and A.I. at Carnegie Mellon University. "There are increasing cyberattacks we're dealing with. We have bigger, more complex problems with all the issues arising out of the explosion of the Web and all of our big data. We're already using A.I. It's already there. I think A.I. in the enterprise is going to start to cascade." ... "