Bill Gates Company to Mine Rare Earth Metals Using Machine Learning

Bill Gates to Mine Rare Earth Metals Using Machine Learning 

Bill Gates’ venture firm, with backing from Jeff Bezos and Jack Ma, just minted a $1 billion A.I. unicorn that uses machine learning for mining rare earth metals crucial for EVs

Mary Meeker and Andreessen Horowitz were also involved in the $250 million funding round.

By Chris Morris in Fortune

June 20, 2023 10:55 AM EDT

Bill Gates’ Breakthrough Energy Ventures is one of the lead investors in a new funding round for KoBold Metals. Jordan Vonderhaar—Bloomberg/Getty Images

KoBold Metals, a Berkeley-based mining company that uses artificial intelligence to mine for rare earth elements, has closed out a funding round that crowns it an A.I. “unicorn” worth $1 billion and ropes in some of the biggest investors in the world.

The $200 million round, reported in the Wall Street Journal, included money from Breakthrough Energy Ventures, a Bill Gates–backed venture capital firm that invests in clean energy companies on behalf of people including Jeff Bezos and Jack Ma.

Other investors included the massively influential VC firm Andreessen Horowitz and Bond Capital, cofounded by Mary Meeker, a longtime VC known for spotting big-picture trends early and whose annual Internet Trends Report is one of the most highly anticipated among tech investors.

The raise brings KoBold’s valuation to more than $1 billion. The company’s last funding round raised $192.5 million in February 2022. Breakthrough Energy Ventures and Bond were both part of that earlier round.

KoBold, founded in 2018, uses machine learning to look for deposits of metals that are crucial to the construction of batteries for electric vehicles—specifically lithium, nickel, cobalt, and copper.

The company has more than 60 projects across three continents, in which it is currently investing $100 million annually. It also has an extensive research and development budget devoted to A.I.’s use in modeling the earth’s subsurface and exploring depths beyond the reach of conventional techniques.

“Our proprietary A.I. tools build on a concept we call efficacy of information (EOI), enabl[ing] KoBold to determine which data to collect at each exploration step, to maximize uncertainty reduction,” the company said on its website.

It might take longer to reach the metals, but the drilling will be precision-focused, the company says. The recent purchase of a copper deposit in Zambia, for example, is expected to take eight years to yield results.

Demand for lithium, nickel, cobalt, and copper is on the rise. By 2050, KoBold predicts, there will be a $12 trillion gap between supply and demand.

Organizing Unstructured Data with Machine Learning

Considering all the data and its uses. 

Organizing Unstructured Data

By Esther Shein

Commissioned by CACM Staff, June 8, 2023

Vector databases are efficient for conducting similarity searches, and they are scalable and flexible, but high-dimensional vectors can be computationally expensive, according to Apple's Huaping Gu.

Vector database company Pinecone in April secured $100 million in venture capital (VC) funding in a $750-million valuation. Other vector database startups have also recently raised millions from VCs, including Chroma, Weviate, and Qdrant. This begs the question: what exactly are vector databases, and why are they generating buzz now?

Some 80% to 90% of any organization's data is unstructured, according to analysts' estimates, and databases have gone through many iterations, from Structured Query Language/SQL databases(in which data is structured in a collection of tables) and relational databases (which focus on the relationship between stored data elements) to NoSQL databases (in which data is stored and retrieved in different structures without using rows and columns). NoSQL was triggered by the advent of Web 2.0 in the early 2000s.

Those traditional databases were not adequately equipped to analyze unstructured data, especially in real time. Now, with artificial intelligence (AI) gaining momentum, vector databases have emerged for use in machine learning applications. A vector is a high-dimensional array of data in which  each dimension is a number.

Explains Charles Xie, CEO and founder of vector database company Zilliz and the Linux Foundation's Milvus Project, "Vectors are important because when you're talking about pictures or images or video, they are the numerical representation of unstructured data that can be easily processed by a machine,''

This is where the use of machine learning models to turn unstructured data into floating point values, or vector embeddings, is key. In contrast, those unstructured images, pictures, and videos are time-consuming and a challenge to classify manually in relational databases. As an example, it took 25,000 people (curators) to label the now-famous ImageNet dataset, Xie says.

Once the data is in a machine-readable format, relational databases store and search across structured table-based data, Xie says. However, unlike structured data, there is no easy way to store and efficiently search large amounts of unstructured data within a relational database.

For example, quickly searching for similar shoes, given a collection of shoe pictures from various angles, would be impossible in a relational database since understanding shoe size, style, heel type, color, etc., purely from the image's raw pixel values is difficult, observes Chris Churilo, vice president of marketing at Zilliz."So we want to turn to a machine to do that for us," using models "that are going to spit out a numerical representation of this content'' that are embeddings or vectors, she says. "The cool thing about having this numerical representation is, now I can ask the machine to find [something] that's similar by basically comparing these numbers against each other."

The machine can do that pretty accurately, Churilo says.  ... '

Successfully Deploying Machine Learning

Successfully Deploying Machine Learning

By focusing on data, value propositions, and talent, enterprises can get ready to take the next step beyond experimentation.

By MIT Technology Review Insights archive page

June 12, 2023

In association with JPMorgan Chase & Co.

After decades of research and development, mostly confined to academia and projects in large organizations, artificial intelligence (AI) and machine learning (ML) are advancing into every corner of the modern enterprise, from chatbots to tractors, and financial markets to medical research. But companies are struggling to move from individual use cases to organization-wide adoption for several reasons, including inadequate or inappropriate data, talent gaps, unclear value propositions, and concerns about risk and responsibility.

Successfully deploying machine learning

Download the report

This MIT Technology Review Insights report, commissioned by and produced in association with with JPMorgan Chase, draws from a survey of 300 executives and interviews with seven experts from finance, health care, academia, and technology to chart elements that are enablers and barriers on the journey to AI/ML deployment.

The following are the report’s key findings:

Businesses buy into AI/ML, but struggle to scale across the organization. The vast majority (93%) of respondents have several experimental or in-use AI/ML projects, with larger companies likely to have greater deployment. A majority (82%) say ML investment will increase during the next 18 months, and closely tie AI and ML to revenue goals. Yet scaling is a major challenge, as is hiring skilled workers, finding appropriate use cases, and showing value.

Deployment success requires a talent and skills strategy. The challenge goes further than attracting core data scientists. Firms need hybrid and translator talent to guide AI/ML design, testing, and governance, and a workforce strategy to ensure all users play a role in technology development. Competitive companies should offer clear opportunities, progression, and impacts for workers that set them apart. For the broader workforce, upskilling and engagement are key to support AI/ML innovations.

Centers of excellence (CoE) provide a foundation for broad deployment, balancing technology-sharing with tailored solutions. Companies with mature capabilities, usually larger companies, tend to develop systems in-house. A CoE provides a hub-and-spoke model, with core ML consulting across divisions to develop widely deployable solutions alongside bespoke tools. ML teams should be incentivized to stay abreast of rapidly evolving AI/ML data science developments. .... ' 

Exploring Material Interiors with Deep Learning

Previously Mentioned, now in more detail..

Machine learning & AI

Deep-learning system explores materials' interiors from the outside   by David L. Chandler , Massachusetts Institute of Technology 

Maybe you can't tell a book from its cover, but according to researchers at MIT you may now be able to do the equivalent for materials of all sorts, from an airplane part to a medical implant. Their new approach allows engineers to figure out what's going on inside simply by observing properties of the material's surface.

The team used a type of machine learning known as deep learning to compare a large set of simulated data about materials' external force fields and the corresponding internal structure, and used that to generate a system that could make reliable predictions of the interior from the surface data.

The results have been published in the journal Advanced Materials, in a paper by doctoral student Zhenze Yang and professor of civil and environmental engineering Markus Buehler.

"It's a very common problem in engineering," Buehler explains. "If you have a piece of material—maybe it's a door on a car or a piece of an airplane—and you want to know what's inside that material, you might measure the strains on the surface by taking images and computing how much deformation you have. But you can't really look inside the material. The only way you can do that is by cutting it and then looking inside and seeing if there's any kind of damage in there."

It's also possible to use X-rays and other techniques, but these tend to be expensive and require bulky equipment, he says. "So, what we have done is basically ask the question: Can we develop an AI algorithm that could look at what's going on at the surface, which we can easily see either using a microscope or taking a photo, or maybe just measuring things on the surface of the material, and then trying to figure out what's actually going on inside?" That inside information might include any damages, cracks, or stresses in the material, or details of its internal microstructure.

The same kind of questions can apply to biological tissues as well, he adds. "Is there disease in there, or some kind of growth or changes in the tissue?" The aim was to develop a system that could answer these kinds of questions in a completely noninvasive way.

Achieving that goal involved addressing complexities including the fact that "many such problems have multiple solutions," Buehler says. For example, many different internal configurations might exhibit the same surface properties. To deal with that ambiguity, "we have created methods that can give us all the possibilities, all the options, basically, that might result in this particular [surface] scenario."

The technique they developed involved training an AI model using vast amounts of data about surface measurements and the interior properties associated with them. This included not only uniform materials but also ones with different materials in combination. "Some new airplanes are made out of composites, so they have deliberate designs of having different phases," Buehler says. "And of course, in biology as well, any kind of biological material will be made out of multiple components and they have very different properties, like in bone, where you have very soft protein, and then you have very rigid mineral substances."  ... ' 

About Max Tegmark, Swedish-American physicist, cosmologist and machine learning researcher

Max Tegmark Interviewed by 'Lex Fridman Podcast #155' :  (Very good) 

https://www.youtube.com/watch?v=RL4j4KPwNGM

https://en.wikipedia.org/wiki/Max_Tegmark

Max Tegmark.jpg

Born 5 May 1967 (age 56)

Stockholm, Sweden

Nationality Swedish  American

Alma mater Royal Institute of Technology

UC Berkeley

Scientific career

Fields Cosmology, physics, machine learning

Institution MIT

Future of Life Institute    https://futureoflife.org/ 

Doctoral advisor Joseph Silk

Max Erik Tegmark[1] (born 5 May 1967) is a Swedish-American physicist, cosmologist and machine learning researcher. He is a professor at the Massachusetts Institute of Technology and the president of the Future of Life Institute.  https://futureoflife.org/  He is also a scientific director at the Foundational Questions Institute and a supporter of the effective altruism movement.[2][3][4][5]  ... 

Safe and Secure Abstractions for Secure Machine Learning

 Intro

Technical Perspective: Beautiful Symbolic Abstractions for Safe and Secure Machine Learning

By Martin Vechev ..   Intro 

Communications of the ACM, February 2023, Vol. 66 No. 2, Page 104     10.1145/3576893

Over the last decade, machine learning has revolutionized entire areas of science ranging from drug discovery to autonomous driving, to medical diagnostics, to natural language processing and many others. Despite this impressive progress, it has become increasingly evident that modern machine learning models suffer from several issues which, if not resolved, could prevent their widespread adoption. Example challenges include lack of robustness guarantees to slight distribution shifts, reinforcing unfair bias present in training data, leakage of sensitive information through the model, and others.

Addressing these issues by inventing new methods and tools for establishing that machine learning models enjoy certain desirable guarantees, is critical, especially for domains where safety and security are paramount. Indeed, over the last few years there has been substantial research progress in new techniques aiming to address the above issues with most work so far focusing on perturbations applied to inputs of the model. For instance, the community has developed novel verification methods for proving that a model always classifies a sample (for example, an image) to the same label regardless of certain transformations (for example, an arbitrary rotation of up to five degrees). New sophisticated methods are constantly being invented targeting different properties, different types of guarantees (probabilistic, deterministic) and application domains (for example, natural language or visual perception).  ... ' 

Anomaly Detection, Supervised or Unsupervised. A Space We worked with Key Data

This kind of data is everywhere.  

Unsupervised and semi-supervised anomaly detection with data-centric ML

February 08, 2023  In Googleblog

Posted by Jinsung Yoon and Sercan O. Arik, Research Scientists, Google Research, Cloud AI Team

Anomaly detection (AD), the task of distinguishing anomalies from normal data, plays a vital role in many real-world applications, such as detecting faulty products from vision sensors in manufacturing, fraudulent behaviors in financial transactions, or network security threats. Depending on the availability of the type of data — negative (normal) vs. positive (anomalous) and the availability of their labels — the task of AD involves different challenges.

(a) Fully supervised anomaly detection, (b) normal-only anomaly detection, (c, d, e) semi-supervised anomaly detection, (f) unsupervised anomaly detection.

While most previous works were shown to be effective for cases with fully-labeled data (either (a) or (b) in the above figure), such settings are less common in practice because labels are particularly tedious to obtain. In most scenarios users have a limited labeling budget, and sometimes there aren’t even any labeled samples during training. Furthermore, even when labeled data are available, there could be biases in the way samples are labeled, causing distribution differences. Such real-world data challenges limit the achievable accuracy of prior methods in detecting anomalies.

This post covers two of our recent papers on AD, published in Transactions on Machine Learning Research (TMLR), that address the above challenges in unsupervised and semi-supervised settings. Using data-centric approaches, we show state-of-the-art results in both. In “Self-supervised, Refine, Repeat: Improving Unsupervised Anomaly Detection”, we propose a novel unsupervised AD framework that relies on the principles of self-supervised learning without labels and iterative data refinement based on the agreement of one-class classifier (OCC) outputs. In “SPADE: Semi-supervised Anomaly Detection under Distribution Mismatch”, we propose a novel semi-supervised AD framework that yields robust performance even under distribution mismatch with limited labeled samples. ... '

Securing Machine Learning is hard

 Via Schneier, with the usual insightful comments.

This is really interesting research from IEEE a few months ago:

Abstract: Given the computational cost and technical expertise required to train machine learning models, users may delegate the task of learning to a service provider. Delegation of learning has clear benefits, and at the same time raises serious concerns of trust. This work studies possible abuses of power by untrusted learners. We show how a malicious learner can plant an undetectable backdoor into a classifier. On the surface, such a backdoored classifier behaves normally, but in reality, the learner maintains a mechanism for changing the classification of any input, with only a slight perturbation. Importantly, without the appropriate “backdoor key,” the mechanism is hidden and cannot be detected by any computationally-bounded observer. We demonstrate two frameworks for planting undetectable backdoors, with incomparable guarantees. .... '

Start with the Basics for Machine Learning

Pretraining and related topics. 

Machines Learn Better if We Teach Them the Basics

A wave of research improves reinforcement learning algorithms by pre-training them as if they were human.      By Max G Levy     In QuantaMagazine

Imagine that your neighbor calls to ask a favor: Could you please feed their pet rabbit some carrot slices? Easy enough, you’d think. You can imagine their kitchen, even if you’ve never been there — carrots in a fridge, a drawer holding various knives. It’s abstract knowledge: You don’t know what your neighbor’s carrots and knives look like exactly, but you won’t take a spoon to a cucumber.

Artificial intelligence programs can’t compete. What seems to you like an easy task is a huge undertaking for current algorithms.

An AI-trained robot can find a specified knife and carrot hiding in a familiar kitchen, but in a different kitchen it will lack the abstract skills to succeed. “They don’t generalize to new environments,” said Victor Zhong, a graduate student in computer science at the University of Washington. The machine fails because there’s simply too much to learn, and too vast a space to explore.

The problem is that these robots — and AI agents in general — don’t have a foundation of concepts to build on. They don’t know what a knife or a carrot really is, much less how to open a drawer, choose one and cut slices. This limitation is due in part to the fact that many advanced AI systems get trained with a method called reinforcement learning that’s essentially self-education through trial and error. AI agents trained with reinforcement learning can execute the job they were trained to do very well, in the environment they were trained to do it in. But change the job or the environment, and these systems will often fail.

ARTIFICIAL INTELLIGENCE

Computers Evolve a New Path Toward Human Intelligence   NOVEMBER 6, 2019

To get around this limitation, computer scientists have begun to teach machines important concepts before setting them loose. It’s like reading a manual before using new software: You could try to explore without it, but you’ll learn far faster with it. “Humans learn through a combination of both doing and reading,” said Karthik Narasimhan, a computer scientist at Princeton University. “We want machines to do the same.”

New work from Zhong and others shows that priming a learning model in this way can supercharge learning in simulated environments, both online and in the real world with robots. And it doesn’t just make algorithms learn faster — it guides them toward skills they’d otherwise never learn. Researchers want these agents to become generalists, capable of learning anything from chess to shopping to cleaning. And as demonstrations become more practical, scientists think this approach might even change how humans can interact with robots.

“It’s been a pretty big breakthrough,” said Brian Ichter, a research scientist in robotics at Google. “It’s pretty unimaginable how far it’s come in a year and a half.”

Sparse Rewards

At first glance, machine learning has already been remarkably successful. Most models typically use reinforcement learning, where algorithms learn by getting rewards. They begin totally ignorant, but trial and error eventually becomes trial and triumph. Reinforcement learning agents can easily master simple games.

Consider the video game Snake, where players control a snake that grows longer as it eats digital apples. You want your snake to eat the most apples, stay within the boundaries and avoid running into its increasingly bulky body. Such clear right and wrong outcomes give a well-rewarded machine agent positive feedback, so enough attempts can take it from “noob” to High Score.

But suppose the rules change. Perhaps the same agent must play on a larger grid and in three dimensions. While a human player could adapt quickly, the machine can’t, because of two critical weaknesses. First, the larger space means it takes longer for the snake to stumble upon apples, and learning slows exponentially when rewards become sparse. Second, the new dimension provides a totally new experience, and reinforcement learning struggles to generalize to new challenges .... '

Attacking Machine Learning Systems

 Considerable piece in Schneier

Just now reading, suggested ...

Attacking Machine Learning Systems

The field of machine learning (ML) security—and corresponding adversarial ML—is rapidly advancing as researchers develop sophisticated techniques to perturb, disrupt, or steal the ML model or data. It’s a heady time; because we know so little about the security of these systems, there are many opportunities for new researchers to publish in this field. In many ways, this circumstance reminds me of the cryptanalysis field in the 1990. And there is a lesson in that similarity: the complex mathematical attacks make for good academic papers, but we mustn’t lose sight of the fact that insecure software will be the likely attack vector for most ML systems.

We are amazed by real-world demonstrations of adversarial attacks on ML systems, such as a 3D-printed object that looks like a turtle but is recognized (from any orientation) by the ML system as a gun. Or adding a few stickers that look like smudges to a stop sign so that it is recognized by a state-of-the-art system as a 45 mi/h speed limit sign. But what if, instead, somebody hacked into the system and just switched the labels for “gun” and “turtle” or swapped “stop” and “45 mi/h”? Systems can only match images with human-provided labels, so the software would never notice the switch. That is far easier and will remain a problem even if systems are developed that are robust to those adversarial attacks.  ... ( Much more)

Lucy 4 and Generative AI

 Lucy 4 just brought to my attention in this realm. 

AI News

Lucy 4 is moving ahead with generative AI for knowledge management

By James Bourne | February 3, 2023 | TechForge Media https://www.cloudcomputing-news.net/

Categories: Applications, Machine Learning,

When it comes to workplace bugbears, wasting time fruitlessly searching shared drives for a particular resource has to be up there. Yet would it not be easier to lighten the workload through an answer engine with a sprinkling of generative AI?  

Machine learning software, by definition, is self-learning. As users ask more questions of an AI, and the AI provides answers, feedback loops are developed which help the product get stronger and the return on investment become greater. 

“It’s really cool that a proper AI solution is self-learning,” Scott Litman, founder and chief operating officer of AI-powered answer engine Lucy, explains. “The AI is growing with them. If the AI misses, it’s a teachable moment, and [it] will be smarter tomorrow.” 

With generative AI, the stakes are now so much higher. Generative AI is defined as algorithms which can be used to create new content, from text, to code, to audio. ChatGPT, from OpenAI, has understandably garnered a fleet of headlines because it appears to have opened up a world of possibility for content creation.  

Yet it is not all plain sailing. For one, users have delighted in pointing out the fallibilities of ChatGPT, which is fine – it is always learning after all. But other users have spotted the software’s tendency to make up a response if it is unsure. “The smug confidence with which [the] AI asserts totally incorrect information is striking,” the writer Ted Gioia noted. “A con artist could not do better.” 

Lucy’s job is not to make incorrect assertions, but to ‘liberate corporate knowledge’: put simply, get the right answer to the right person at the right time in seconds, regardless of where that answer lives. Much of this will primarily involve sifting through reams of PDFs, PowerPoints and Word documents and point to the most relevant detail, but this liberation can turn up insights in previously forgotten places, such as video training courses. 

With the recent release of Lucy 4, the next generation of its platform, and Lucy Synopsis, there is a further push towards generative AI – but without the drawbacks. Lucy can not only point a user to an answer, but provide a unique two-to-three sentence summary which directly answers the question. Crucially, as Steve Frederickson, director of product management points out, Lucy’s summations are there solely to help the user, not offer a spurious alternative. 

One of the key elements of Lucy 4, again involving the generative AI element, is expanded integration with Microsoft Teams and Slack, where users can mention Lucy in a chat. This reflects not just greater ease of use for employees, but a wider trend around search.  

“One of the things we realised last year was that, along with the inefficiency of searching, people in some cases have given up on the idea of searching,” explains Litman. The result is that users are more likely to fire out a message on the chat apps than waste time on a frustrating scavenger hunt. “Which sometimes works – human intelligence is a great thing,” says Litman. “But if you’re the subject matter expert answering all the questions, you’re constantly being disrupted.” 

“We come at it from our own perspective – we have a core value of experimentation,” adds Frederickson. “Lucy has always had the tenet of going above and beyond search. We hold ourselves to that higher standard.”   .... .'

Empirical Optimization with Divergent Fixed Point Algorithm – When All Else Fails

 Of Technical Interest, worth signing into:

Empirical Optimization with Divergent Fixed Point Algorithm – When All Else Fails

 DECEMBER 9, 2022 EXPLAINABLE AI FEATURED POSTS MACHINE LEARNING SYNTHETIC DATA VISUALIZATION

Entitled “Empirical Optimization with Divergent Fixed Point Algorithm – When All Else Fails”, the full version in PDF format is accessible in the “Free Books and Articles” section, here. Also discussed in details with Python code in my book “Synthetic Data”, available here.

While the technique discussed here is a last resort solution when all else fails, it is actually more powerful than it seems at first glance. First, it also works in standard cases with “nice” functions. However, there are better methods when the function behaves nicely, taking advantage of the differentiability of the function in question, such as the Newton algorithm (itself a fixed-point iteration). It can be generalized to higher dimensions, though I focus on univariate functions here.

Perhaps the attractive features are the fact that it is simple and intuitive, and quickly leads to a solution despite the absence of convergence. However, it is an empirical method and may require working with different parameter sets to actually find a solution. Still, it can be turned into a black-box solution by automatically testing different parameter configurations. In that respect, I compare it to the empirical elbow rule to detect the number of clusters in unsupervised clustering problems. I also turned the elbow rule into a fully automated black-box procedure, with full details offered in the same book .... '

About the author

Vincent Granville is a pioneering data scientist and machine learning expert, co-founder of Data Science Central (acquired by  TechTarget in 2020), founder of MLTechniques.com, former VC-funded executive, author and patent owner. Vincent’s past corporate experience includes Visa, Wells Fargo, eBay, NBC, Microsoft, and CNET. Vincent is also a former post-doc at Cambridge University, and the National Institute of Statistical Sciences (NISS).  

Vincent published in Journal of Number Theory, Journal of the Royal Statistical Society (Series B), and IEEE Transactions on Pattern Analysis and Machine Intelligence. He is also the author of multiple books, including “Intuitive Machine Learning and Explainable AI”, available here. He lives  in Washington state, and enjoys doing research on spatial stochastic processes, chaotic dynamical systems, experimental math and probabilistic number theory.  

Synthetic Data Can Offer ML Performance Improvements

 Previously mentioned, considering use of idea in test.

MIT News, Adam Zewe, November 3, 2022

A team of researchers at the Massachusetts Institute of Technology (MIT), the MIT-IBM Watson AI Laboratory, and Boston University found synthetic data can improve machine learning (ML) model performance. The researchers amassed the Synthetic Action Pre-training and Transfer dataset of 150,000 video clips to train three ML models on a wide range of human actions. They found the three models outperformed models trained with real video clips on four of six datasets, yielding the highest accuracy for datasets featuring clips with "low scene-object bias." MIT-IBM Watson AI Lab's Rogerio Feris said, "The ultimate goal of our research is to replace real data pretraining with synthetic data pretraining." Feris said that while there is a cost in creating an action in synthetic data, “Once that is done, then you can generate an unlimited number of images or videos by changing the pose, the lighting, etc."  ... ' 

Choosing Best Machine Learning Algorithm

 Simplified example

DataScienceDojo (@Data Science Dojo) Tweeted: 🤔 Here is how to choose the correct machine-learning algorithm!

Source: datasciencebrain

https://twitter.com/DataScienceDojo

#MachineLearning #DataScience https://twitter.com/datasciencedojo/status/1588715224605106176?s=51&t=e1T9N4c2nhYKMsgxSF1GFQ

What is AI? Useful basic intro Discussion

A nice piece from Computing practitioner David Bombal, his Youtube link below:

https://www.youtube.com/watch?v=PH9RQ6Yx75c&t=47s

Computerphile: David Bombal     Many related links below

If you are into cybersecurity or any other field, you probably want to learn about AI and ML. They can really help your resume and help you increase the $$$ you earn.

AI just become Sentient? And will it take your job? Or is AI just a fantastic opportunity for you to get a better job? In this interview with Dr Michael Pound we discuss hype vs reality and get a quick start guide on how to learn AI. 

// MENU //

00:00 - Coming Up

00:45 - Intro

01:10 - Michael Pound introduction

02:49 - Will AI take our jobs?

04:55 - What is LaMDA?

08:38 - Can Python functions get lonely?

11:26 - The definition of "sentience"

11:59 - AI vs Machine Learning

18:48 - Neural Networks

19:49 - Malware example

21:59 - Stochastic Gradient Descent

22:30 - Supervised learning

23:45 - Unsupervised learning

26:03 - Reinforcement learning

27:35 - Are the robots taking over?

30:14 - What is AI really good at?

33:28 - Definition of Deep Learning

35:37 - Neural Networks

36:53 - What to learn

40:50 - Using PyTorch

43:52 - Google colab

44:48 - Study recommendations

46:16 - The demand for AI skills

48:15 - Teaching cyber security 

50:06 - Final Advice

55:09 - Conclusion

// Video mentions //

ComputerPhile (lambda is not sentient): https://youtu.be/iBouACLc-hw

Data Analysis Playlist: https://www.youtube.com/watch?v=NxYEz...

Neural Networks Playlist: https://www.youtube.com/watch?v=py5by...

Computer Vision Playlist: https://www.youtube.com/watch?v=C_zFh...

// BOOK //

Deep learning by Ian Goodfellow, Yoshua Bengio and Aaron Courville: https://amzn.to/3vmu4LP

// COURSE //

AI For Everyone by Andrew Ng: https://www.coursera.org/learn/ai-for... 

Thinking, Detecting Bots

Home/News/Bot Hunting Is All About the Vibes 

ACM NEWS

Bot Hunting Is All About the Vibes

By Wired, October 4, 2022

Christopher Bouzy is trying to stay ahead of the bots. As the person behind Bot Sentinel, a popular bot-detection system, he and his team continuously update their machine learning models out of fear that they will get "stale." The task? Sorting 3.2 million tweets from suspended accounts into two folders: "Bot" or "Not."

To detect bots, Bot Sentinel's models must first learn what problematic behavior is through exposure to data. And by providing the model with tweets in two distinct categories—bot or not a bot—Bouzy's model can calibrate itself and allegedly find the very essence of what, he thinks, makes a tweet problematic.

Training data is the heart of any machine learning model. In the burgeoning field of bot detection, how bot hunters define and label tweets determines the way their systems interpret and classify bot-like behavior. According to experts, this can be more of an art than a science. "At the end of the day, it is about a vibe when you are doing the labeling," Bouzy says. "It's not just about the words in the tweet, context matters." ...

Before anyone can hunt bots, they need to figure out what a bot is—and that answer changes depending on who you ask. .... 

In Wired

Optimizing Fluid Mixing with Machine Learning

 Interaction between RL and Markov decision method is interesting 

Optimizing Fluid Mixing with Machine Learning

Tokyo University of Science (Japan)

August 29, 2022

Researchers in Japan have proposed a machine learning-based approach for optimizing fluid mixing for laminar flows. The researchers used reinforcement learning (RL), in which intelligent agents perform actions in an environment to maximize the cumulative reward. The team addressed RL's inefficiencies in dealing with systems involving high-dimensional state spaces by describing fluid motion using only a single parameter. Researchers used the Markov decision process to formulate the RL algorithm, and the Tokyo University of Science's Masanobu Inubushi said the program "identified an effective flow control, which culminated in an exponentially fast mixing without any prior knowledge." The RL method also enabled effective transfer learning of the trained "mixer," significantly reducing its time and training cost.

Improving Machine Learning Outcomes

Published in Towards Data Science   By John Hawkins, Chief Scientist at Playground XYZ. Computer scientist, open source developer and the author of getting-data-science-done.com

Improving Machine Learning Outcomes

Focusing on Framing, Timing, and Targets — Introduction In order to build successful machine learning solutions, there are certain fundamental ideas that everyone involved needs to understand. In this blog post, we look at three key early stages of the design process that managers can focus on to ensure that the project is headed toward a successful outcome. …

Machine Learning 

Introduction

In order to build successful machine learning solutions, there are certain fundamental ideas that everyone involved needs to understand. In this blog post, we look at three key early stages of the design process that managers can focus on to ensure that the project is headed toward a successful outcome.

This post presumes the reader already understands distinctions in machine learning such as supervised and unsupervised models, training and testing stages, and the overall machine learning lifecycle. Returning to the earliest stage of defining the business problem, we focus our attention on three key objectives.

Three Key Objectives: Framing, Timing, & Target

Each of these objectives are introduced to some extent in data science training. However, quite often they do not receive the emphasis they deserve. In part this is because, on the surface, they seem obvious: frame the problem, decide when predictions will be made, and determine a target variable for the model to learn from.

These are three major levers a project stakeholder can utilize to influence the likelihood of project success. For this reason, it’s important to dive deeper on each one in order to surface some of the subtleties that can differentiate between getting a model that solves your business problem, or not.  ..... ' 

Interpretable Machine Learning and Diagnosis

Well covered in ACM pieces of late, useful takes. 

ACM PRACTICE

Interpretable Machine Learning: Moving from Mythos to Diagnostics

By Valerie Chen, Jeffrey Li, Joon Sik Kim, Gregory Plumb, Ameet Talwalkar

Communications of the ACM, August 2022, Vol. 65 No. 8, Pages 43-50  10.1145/3546036

The emergence of machine learning as a society-changing technology in the past decade has triggered concerns about people's inability to understand the reasoning of increasingly complex models. The field of interpretable machine learning (IML) grew out of these concerns, with the goal of empowering various stakeholders to tackle use cases, such as building trust in models, performing model debugging, and generally informing real human decision-making.7,10,17

Yet despite the flurry of IML methodological development over the past several years, a stark disconnect characterizes the current overall approach. As shown in Figure 1, IML researchers develop methods that typically optimize for diverse but narrow technical objectives, yet their claimed use cases for consumers remain broad and often underspecified. Echoing similar critiques about the field,17 it has thus remained difficult to evaluate these claims sufficiently and to translate methodological advances into widespread practical impact. .... '

We are Training Much Faster Now

Faster, better training is here outlined and described. 

We’re Training AI Twice as Fast This Year as Last New MLPerf rankings show training times plunging     By SAMUEL K. MOORE    in IEEE Spectrum

According to the best measures we’ve got, a set of benchmarks called MLPerf, machine-learning systems can be trained nearly twice as quickly as they could last year. It’s a figure that outstrips Moore’s Law, but also one we’ve come to expect. Most of the gain is thanks to software and systems innovations, but this year also gave the first peek at what some new processors, notably from Graphcore and Intel subsidiary Habana Labs, can do.

The once-crippling time it took to train a neural network to do its task is the problem that launched startups like Cerebras and SambaNova and drove companies like Google to develop machine-learning accelerator chips in house. But the new MLPerf data shows that training time for standard neural networks has gotten a lot less taxing in a short period of time. And that speedup has come from much more than just the advance of Moore’s Law.

This capability has only incentivized machine-learning experts to dream big. So the size of new neural networks continues to outpace computing power.

Called by some “the Olympics of machine learning,” MLPerf consists of eight benchmark tests: image recognition, medical-imaging segmentation, two versions of object detection, speech recognition, natural-language processing, recommendation, and a form of gameplay called reinforcement learning. (One of the object-detection benchmarks was updated for this round to a neural net that is closer to the state of the art.) Computers and software from 21 companies and institutions compete on any or all of the tests. This time around, officially called MLPerf Training 2.0, they collectively submitted 250 results.

Very few commercial and cloud systems were tested on all eight, but Nvidia director of product development for accelerated computing Shar Narasimhan gave an interesting example of why systems should be able to handle such breadth: Imagine a person with a smartphone snapping a photo of a flower and asking the phone: “What kind of flower is this?” It seems like a single request, but answering it would likely involve 10 different machine-learning models, several of which are represented in MLPerf.

To give a taste of the data, for each benchmark we’ve listed the fastest results for commercially available computers and cloud offerings (Microsoft Azure and Google Cloud) by how many machine-learning accelerators (usually GPUs) were involved. Keep in mind that some of these will be a category of one. For instance, there really aren’t that many places that can devote thousands of GPUs to a task. Likewise, there are some benchmarks where systems beat their nearest competitor by a matter of seconds or where five or more entries landed within a few minutes of each other. So if you’re curious about the nuances of AI performance, check out the complete list.  .... '   (much more at link)