AI self-play for Algorithm Design

AI and Puzzles..

AI self-play for algorithm design   in Microsoft.com

Published May 2, 2023

By Adam Tauman Kalai , Senior Principal Researcher  Patrick Haluptzok , AI Research Associate

A self-play pipeline for a language model (LM) to improve itself in a fully automatic manner. First, the LM generates novel puzzles based on a training set of handwritten puzzles. Then, the LM attempts to solve each of these puzzles 100 times. In Step 3, the computer (specifically a Python interpreter) filters the candidate solutions for correctness. Finally, the LM is improved by further training on these verified correct solutions to synthetic puzzles, and the process repeats. This process leads to significant improvements as measured on held-out test puzzles, which were also handwritten.

Efficient algorithms are crucial for many purposes, including reducing energy consumption in digital devices. While humans outperform AI systems at designing such algorithms, we show how to improve AI programming abilities using self-play, a technique that has helped AI systems dominate in games such as chess and Go.

Designing fast and accurate algorithms requires high-level abstract reasoning, which remains difficult for AI systems. Our approach involves having the AI design and solve its own programming challenges, enabling practice on millions of artificial challenges and exploration of problem types not found in public repositories. We detail our work in a new paper, “Language Models Can Teach Themselves to Program Better,” which we’re presenting at the 2023 International Conference on Learning Representations (ICLR).  ... '

Designing with DNA

Like the idea, examining the approach further.

Designing with DNA    By Duke University, December 28, 2022

An open-source software program developed by researchers at Duke and other universities lets users take drawings or digital models of rounded shapes and turn them into 3-D structures made of DNA.

While researchers have been experimenting with DNA as a construction material since the 1980s, designing structures with curved surfaces like those found in nature has been tricky. Duke Ph.D. student Dan Fu is part of a team that developed software called DNAxiS, which relies on a previously described way to build with DNA by coiling a long DNA double helix into concentric rings that stack on each other. DNAxiS lets users design shapes automatically, using algorithms to determine where to place short DNA "staples" to join the longer DNA rings together and hold the shape in place.

"If there are too few, or if they're in the wrong position, the structure won't form correctly," Fu says. "Before our software, the curvature of the shapes made this an especially difficult problem."

From Duke University

View Full Article   

New Computing Architecture

At a school I attended, worth a look.   Technical

Researchers at the University of Pennsylvania Propose a New Computing Architecture Ideal for Artificial Intelligence (AI)

By Khushboo Gupta- October 5, 2022

Conventional computing architectures severely constrain artificial intelligence’s ability to improve technology. In traditional models, memory storage and computing occur in separate areas of the machine. This is why data must be transported from its storage area to a CPU or GPU for processing. The most significant disadvantage of this design is that this movement takes time, which reduces the performance of even the most potent processing units available. There is no avoiding lag when compute performance exceeds memory transfer. These delays become a severe issue when dealing with the massive amounts of data required for machine learning and AI applications. 

Researchers have focused on hardware innovation to achieve the necessary increases in speed, agility, and energy efficiency as AI software advances in sophistication and the rise of the sensor-heavy Internet of Things produces larger datasets. A team of researchers from the University of Pennsylvania’s School of Engineering and Applied Science, in collaboration with researchers from Sandia National Laboratories and Brookhaven National Laboratory, have created a new computing architecture based on compute-in-memory (CIM), which is ideal for AI. Processing and storage take place simultaneously in CIM systems, which helps to reduce energy consumption and eliminate transfer time. The new CIM design from the team stands out for containing no transistors. This design is specifically adapted to how Big Data applications have changed how computing works today.

Transistors limit the speed at which data may be accessed, even in a compute-in-memory architecture. They utilize more time, space, and energy than is ideal for AI applications since they require much wire in a chip’s overall circuitry. The transistor-free design by the team is distinctive since it is straightforward, quick, and uses less energy. The researchers clearly emphasize that the advancement is not limited to circuit-level design. Their earlier materials science research on a semiconductor known as scandium-alloyed aluminum nitride (AlScN) was the foundation for the new computing architecture. Ferroelectric switching is possible with AlScN, making it faster and more energy-efficient than other nonvolatile memory components. Another crucial feature is the material’s ability to be deposited at temperatures low enough to work with silicon foundries. This makes it possible for the architecture to be space-efficient, which is crucial for small chip designs. .... '    (much more, Computing Technical) 

Redesigning Streets with AI

Worked with  civil engineers early in my career who worked with this kind of designing, fascinating approach

Think Your Street Needs a Redesign? Ask an AI

Bloomberg, Marie Patino, July 30, 2022

Brooklyn-based artist and former activist Zach Katz is using artificial intelligence (AI) laboratory OpenAI's DALL-E 2 system to redesign streets to be more friendly to pedestrians and bikes. DALL-E 2, the second iteration of the DALL-E neural network, can convert text into visual features and display how these features could relate to one another, thanks to a massive training dataset. Users can input text queries and request specific styles and features from the AI, although results can be biased against women and people of color due to a lack of diversity in its training data. Katz posts images of DALL-E-reimagined streets on Twitter, and has a backlog of requests for such reimaginings from around the world.  .... 

Chip Scans Possible

Chips scans possible to reveal designs

X-Ray Tech Reveals Chip Design    in IEEE

No trade secret or hardware trojan can hide from ptychographic X-ray laminography

By  ANTHONY F.J. LEVI GABRIEL AEPPLI30 APR 2022  

HEN YOU’RE BAKING a cake, it’s hard to know when the inside is in the state you want it to be. The same is true—with much higher stakes—for microelectronic chips: How can engineers confirm that what’s inside has truly met the intent of the designers? How can a semiconductor design company tell whether its intellectual property was stolen? Much more worrisome, how can anyone be sure a kill switch or some other hardware trojan hasn’t been secretly inserted?

Today, that probing is done by grinding away each of the chip’s many layers and inspecting them using an electron microscope. It’s slow going and, of course, destructive, making this approach hardly satisfactory for anybody.

One of us (Levi) works with semiconductors and the other (Aeppli) with X-rays. So, after pondering this problem, we considered using X-rays to nondestructively image chips. You’d need to go beyond the resolution used in medical X-ray scanners. But it was clear to us that the needed resolution was possible. At that moment, what we’ve been calling the “chip scan” project was born.  ...'

Virtual Chemical Reaction Design

ew approaches to chemical design

Chemical Reaction Design Goes Virtual

Hokkaido University (Japan),March 14, 2022

A virtual ligand-assisted (VLA) screening method developed by researchers at Japan's Hokkaido University could reduce time spent on trial and error during transition metal catalyst development. The VLA screening method surveys a wide range of values for different properties to determine the most promising features of ligands, which are molecules bonded to the central metal atom of a catalyst. Virtual ligands mimic the presence of real ligands, but rather than being described by numerous individual constituent atoms, they are described using only their steric (space-filling) and electronic properties. Researchers then generate a contour map detailing the combination of steric and electronic effects a ligand should have to best catalyze a specific reaction, so chemists only must test real ligands meeting this criteria. "As the VLA screening can be conducted in silico, it would save a lot of time and resources in the lab," said Hokkaido's Satoshi Maeda.  ... '

AI Changing Chip Design

 Seeking better chips. 

Q&A: Here’s How AI Will Change Chip Design Artificial intelligence’s promise and potential for the semiconductor industry   By RINA DIANE CABALLAR08 FEB 2022  in IEEE Spectrum

The end of Moore’s Law is looming. Engineers and designers can do only so much to miniaturize transistors and pack as many of them as possible into chips. So they’re turning to other approaches to chip design, incorporating technologies like AI into the process.

Samsung, for instance, is adding AI to its memory chips to enable processing in memory, thereby saving energy and speeding up machine learning. Speaking of speed, Google’s TPU V4 AI chip has doubled its processing power compared with that of its previous version.

But AI holds still more promise and potential for the semiconductor industry. To better understand how AI is set to revolutionize chip design, we spoke with Heather Gorr, senior product manager for MathWorks’ MATLAB platform.

How is AI currently being used to design the next generation of chips?

Heather Gorr: AI is such an important technology because it’s involved in most parts of the cycle, including the design and manufacturing process. There’s a lot of important applications here, even in the general process engineering where we want to optimize things. I think defect detection is a big one at all phases of the process, especially in manufacturing. But even thinking ahead in the design process, [AI now plays a significant role] when you’re designing the light and the sensors and all the different components. There’s a lot of anomaly detection and fault mitigation that you really want to consider.

Then, thinking about the logistical modeling that you see in any industry, there is always planned downtime that you want to mitigate; but you also end up having unplanned downtime. So, looking back at that historical data of when you’ve had those moments where maybe it took a bit longer than expected to manufacture something, you can take a look at all of that data and use AI to try to identify the proximate cause or to see something that might jump out even in the processing and design phases. We think of AI oftentimes as a predictive tool, or as a robot doing something, but a lot of times you get a lot of insight from the data through AI. ......' 

A New Era for Mechanical CAD

Considerable piece.   We actively worked with computer aided design, and used sketchpad too.  Had always thought there would be better models for 'assistance' for its use.  Early on we examined automated efficiency of design measures.    And a means to more readily integrate designs with existing contexts and uses.  Technical. 

A New Era for Mechanical CAD

Time to move forward from decades-old design

Jessie Frazelle  in Queue.acm.org

CAD (computer-aided design) has been around since the 1950s. The first graphical CAD program, called Sketchpad, came out of MIT [designworldonline.com]. Since then, CAD has become essential to designing and manufacturing hardware products. Today, there are multiple types of CAD. This column focuses on mechanical CAD, used for mechanical engineering.

Digging into the history of computer graphics reveals some interesting connections between the most ambitious and notorious engineers. Ivan Sutherland, who won the Turing Award for Sketchpad in 1988, had Edwin Catmull as a student. Catmull and Pat Hanrahan won the Turing award for their contributions to computer graphics in 2019. This included their work at Pixar building RenderMan [pixar.com], which was licensed to other filmmakers. This led to innovations in hardware, software, and GPUs. Without these innovators, there would be no mechanical CAD, nor would animated films be as sophisticated as they are today. There wouldn't even be GPUs.

Modeling geometries has evolved greatly over time. Solids were first modeled as wireframes by representing the object by its edges, line curves, and vertices. This evolved into surface representation using faces, surfaces, edges, and vertices. Surface representation is valuable in robot path planning as well. Wireframe and surface representation contains only geometrical data. Today, modeling includes topological information to describe how the object is bounded and connected, and to describe its neighborhood. (A neighborhood of a point consists of a set of points containing that point where one can move some distance in any direction away from that point without leaving the set.)

OpenCascade, Parasolid, and ACIS are all boundary-representation (B-rep) kernels. A B-rep model is composed of geometry and topology information. The topology information differs depending on the program used. B-rep file formats include STEP (Standard for the Exchange of Product Model Data), IGES (Initial Graphics Exchange Specification), NX's prt, Solid Edge's par and asm, Creo's prt and asm, SolidWorks' sldprt and sldasm, Inventor's ipt and iam, and AutoCAD's dwg.

Visual representation (vis-rep) models tend to be much smaller in data size than B-rep models. This is because they do not contain as much structural or product management information. Vis-rep models are approximations of geometry and are composed of a mass of flat polygons. Vis-rep file formats include obj, STL, 3D XML, 3D PDF, COLLADA, and PLY.

CAD programs tend to use B-rep models, while animation, game development, augmented reality, and virtual reality tend to use vis-rep models. However, the two are interchanged frequently. For example, if you were using a B-rep model for manufacturing but wanted to load it into Apple's ARKit for some animations, you would first convert it to COLLADA, a vis-rep file format. The file should already be a bit smaller from dropping all the CAD data, but if you wanted to make it even smaller, you could tweak the polygon counts on each of the meshes for the various parts.

The tools used to build with today are supported on the shoulders of giants, but a lot could be done to make them even better. At some point, mechanical CAD lost some of its roots of innovation. Let's dive into a few of the problems with the CAD programs that exist today and see how to make them better. .... ' 

AI and Design Thinking

Interesting point.   I say all systems, analytics, processes, management have a 'design problem', they need to be applied properly to provide value.     All statistical and analytical methods were not understood by decision makers,  but still effectively used.  Same with AI.  But do design their use well.

AI has become a design problem

By Mark Rolston, argodesign

Join executive leaders at the Data, Analytics, & Intelligent Automation Summit, presented by Accenture. Watch now!

AI is facing new calls for regulation now that it has emerged from the laboratory and is becoming more widely deployed across our daily lives. The public does not trust the technology. Nor should they. The problems are numerous, from understanding how the models actually work, controlling the data that feeds AI, and addressing growing distrust in those that wield this technology. Even companies themselves are unsure of how to safely and effectively use AI as part of their business.

The bulk of the discussion is currently centered on engineering. It’s understandable — artificial intelligence is a mysterious black box to many people, one assumed to be fixable only through lines of code and better data. But I want to argue that in order to truly understand and begin to master this technology, we must get better at seeing AI as a component of larger systems and design accordingly. Fundamentally, AI is not just a technology problem, it has become a design problem.   ... ' 

Designing VR from Youtube Fails

 Clever idea here, look for fails shown in YouTube to design-in resilience to fails.   An example of a huge source of data to design in such patterns.    Also the associated metadata.  Perhaps with other sources of such broad data as well.

What could possibly go wrong with virtual reality?,   by University of Copenhagen

YouTube is a treasure trove of virtual reality fails: users tripping, colliding into walls and smacking inanimate and animate objects. By investigating these "VR fails' on YouTube, researchers at the University of Copenhagen have sought to learn more about when and why things go sideways for users and how to improve VR design and experiences so as to avoid accidents.

Millions of YouTube viewers have enjoyed hearty laughs watching others getting hurt using virtual reality—people wearing VR headsets, falling, screaming, crashing into walls and TV sets, or knocking spectators to the floor. Some of us have even been that failing someone. Now, videos of virtual reality mishaps, called "VR Fails," have actually become a field of research.  .... " 

Digitization of Creativity and the Maker Movement.

A considerable piece.    In particular note the connection to the 'Maker Movement',  an approach we were introduced to as early as the 80s.  Of course 3D printing has become a big component.   Software tools to to produce models.    And also elements of AI able to work with us to add design creativity?   

What Can the Maker Movement Teach Us About the Digitization of Creativity?   By Sascha FrSiesike, Frédéric Thiesse, George Kuk

Communications of the ACM, March 2021, Vol. 64 No. 3, Pages 42-45   10.1145/3447524

In recent years, the 'maker movement' has emerged as a social phenomenon driven by novel technological possibilities.1 With the help of inexpensive, yet highly versatile means of production (for example, CNC milling machines, 3D printers) and easy-to-use software tools, makers free themselves from their traditional role as passive consumers and evolve into innovators and producers. Although the act of physical production seems to be at the center of the movement, a large part of the creative work takes place in the online sphere. These digital activities and their outcomes provide a rich source of information that can be used to gain a more nuanced understanding of how the digitization affects the creative process itself.

Of all the production methods available to makers, 3D printing is probably the most versatile and requires only a limited understanding of the production process. Several 3D design software packages allow even lay people to turn their ideas into printable designs. This combination of flexibility and usability has led to an abundance of 3D object models over the past years, which are shared and jointly refined with the community on digital maker platforms. As part of a multi-year research project on the use of 3D printing by the maker community, we found that the use of these platforms in the creative process blurs the boundaries between the digital and the physical and ultimately changes the way ideas are expressed, curated, and eventually translated into physical reality. In particular, we saw how makers with entirely different backgrounds (for example, HW/SW developers, designers, business and social entrepreneurs) traverse across the startup world, software development, and open online communities, to combine concepts through a novel digitized creative process. ... ' 

Walking Inside Samples

Worked on something similar, positioned for looking at product designs.   Wlaking through thigs can give you ideas about design. 

Software Allows Scientists to 'Walk Inside' Samples

Australian National University  February 3, 2021

The new Drishti software developed by researchers at Australian National University (ANU) lets scientists visualize data in three dimensions and generate lifelike models of objects like fossil samples so they can "zoom in" on smaller details without damaging the original sample. ANU's Yuzhi Hu said, "After we scan the sample, we then have a set of 3D data which can be digitally dissected effectively using our new tool.” The ANU team said the software could be particularly useful for scientific communication and education. ANU's Ajay Limaye said Drishti's current applications include digitally duplicating a mummy sample. The Drishti software is free and available online... '

Color to Avoid Design Failures

Coloring your designs to avoid failures.

Merging Technologies with Color to Avoid Design Failures

Penn State College of Engineering, Miranda Buckheit, January 21, 2021

Pennsylvania State University (Penn State) researchers analyzed machine learning (ML) and image colorization algorithms to avoid design failures. Penn State's Pranav Milind Khanolkar reviewed the use of the ABAQUS additive-manufacturing simulation software, which can pose difficulties because its speed and performance level rely on a computer's hardware processing power. The team deployed ML algorithms to lower the exclusive use of computationally demanding finite element analysis, and accelerate simulations. It then applied image colorization algorithms to material microstructure data, and repurposed programs typically used to add color to monochrome photos. Khanolkar said, "Using intelligent technology to help people and empower their creativity and empathy during the design process is important. These algorithms need lots of computational power and using artificial intelligence ... allows designers to be more creative without impacting production cost."

Remaking the Web

We need a remake.  With caution to include freedom of communication.

 He Created the Web. Now He's Out to Remake the Digital World.

The New York Times,   Steve Lohr,   January 10, 2021

World Wide Web creator and 2016 ACM A.M. Turing Award recipient Tim Berners-Lee hopes to bring the Web back to his original vision of a decentralized marketplace through personal online data stores (pods), with users controlling their data in individual vaults. Berners-Lee co-founded the startup Inrupt, which will charge licensing fees for its commercial software, using Solid open source software along with augmented security, management, and developer tools. Pods are free for users, and Inrupt expects initial sponsors will be trusted organizations. Increasingly intelligent data protection technology is placing competitive pressure on the big tech firms; Georgia Institute of Technology's Peter Swire said this creates "a market opportunity for Tim Berners-Lee's firm and others to offer individuals better ways to control their data."

Digital Product Development from Visualization

And forward to product design and development, from visualization. 

Visualization Adds Transparency to Digital Product Development

Fraunhofer-Gesellschaft (Germany),  January 4, 2021

Visualization software developed by researchers at Germany's Fraunhofer Institute for Computer Graphics Research (IGD) and Austria's Linz Center of Mechatronics could embed transparency within digital product development. The Fraunhofer IGD team applied this tool to the development of electric motors, while its Linz partner produced technical data and mathematically modeled all product parameters. The visualization covers all criteria relevant for motor development, and portrays the interactions of individual parameters, relaying what occurs when any value is changed and the impact on other criteria, in real time. Fraunhofer IGD's Lena Cibulski said, "The software is an ideal choice whenever dealing with many design options and a number of incompatible quality criteria that require a compromise to be made."

Inferring Design Plans from Short Clips

 Currently examining this kind of approach.  Phones are getting these kinds of abilities. 

Researchers design AI that can infer whole floor plans from short video clips   By Kyle Wiggers   @Kyle_L_Wiggers

Floor plans are useful for visualizing spaces, planning routes, and communicating architectural designs. A robot entering a new building, for instance, can use a floor plan to quickly sense the overall layout. Creating floor plans typically requires a full walkthrough so 3D sensors and cameras can capture the entirety of a space. But researchers at Facebook, the University of Texas at Austin, and Carnegie Mellon University are exploring an AI technique that leverages visuals and audio to reconstruct a floor plan from a short video clip.  ...  '

Templates for Computerized Robot Design

Seems a useful approach as we use robot solutions for more applications. Includes both physical designs and control programming.   And Optimization to tasks involved. 

Computer-Aided Creativity in Robot Design

MIT News,  By Daniel Ackerman

Massachusetts Institute of Technology (MIT) researchers have developed a system that simulates and optimizes robot design and control programs. RoboGrammar avoids jumbled structures yielded from arbitrary connections between parts by following rules on component arrangement, or graph grammar, inspired by arthropods. Based on inputs of available components and intended terrain, RoboGrammar defines the problem, drafts possible solutions, and chooses the optimal ones, using graph grammar to design all permutations. Each robot has a Model Predictive Control algorithm to model its movements and select the best design, while a graph heuristic search algorithm iteratively samples and assesses sets of robots, learning which designs execute a specific task better. Said Columbia University’s Hod Lipson, “This work is a crowning achievement in the 25-year quest to automatically design the morphology and control of robots.”

Apple Using Boston Robotics Spot

 Been following various uses ot the 'Spot'  robot.   And how it makes its telepresence felt by its new subjects.    Here  new and different application, working with the construction of a building, apparently during the design and construction phase. 

Meet Apple’s telepresence robot

In APPLE HOLIC  By Jonny Evans, Computerworld

A Boston Dynamics robot called Spot is helping Apple and its architects design a new HQ at an iconic location.

The COVID-19 pandemic has forced every industry to improvise and innovate to build agility and resilience to get through the crisis. Now, a key Apple partner is using telepresence robots as it leads construction of a Apple’s new headquarters in London. .... 

Apple’s billion-dollar Californian headquarters is somewhat under-used at the moment, but the company continues to invest in new centers worldwide as it continues to experience global growth. Apple in the UK has taken over a large part of the globally recognized Battersea Power Station for its new UK headquarters.   ... " 

On IOT Security: Guidelines for Improvement

 As we depend on many new communicating  'things' in our environment, their design and management over time and context must be done better.

IoT Security Is a Mess. These Guidelines Could Help Fix That

By Danny Palmer in ZDNet

New guidelines from the European Union Agency for Cybersecurity (ENISA) specify recommendations for strengthening Internet of Things (IoT) security throughout product development. The Guidelines for Securing the IoT-Secure Supply Chain for IoT urge further integration of cybersecurity expertise within all organizational layers so supply-chain participants can identify potential risks before they become serious. ENISA also recommends adopting "Security by Design" across the IoT development cycle, with emphasis on careful planning and risk management. Moreover, the guidelines advise organizations to nurture better relationships throughout product development and deployment, in order to close security loopholes that may emerge when communication between participants is lacking. ENISA’s Juhan Lepassaar said, "Securing the supply chain of ICT products and services should be a prerequisite for their further adoption, particularly for critical infrastructure and services. Only then can we reap the benefits associated with their widespread deployment, as it happens with IoT.”  ... ' 

Designing the Echo

 Fast Company looks at the design and redesign of the Amazon Echo since 2014.  Thoughtful examination of how complex consumer design works.

Why Amazon radically redesigned the Echo

You may be ordering one for Black Friday. But why is it shaped like that? ...