Machine Learning: Compare and Contrast
February 2, 2022
For machine learning pros looking to choose a framework, Analytics India Magazine examines one of the most prominent options alongside a new contender in, “TensorFlow Vs PyCaret: A Comparison of Machine Learning Frameworks.” The article takes us through an informative juxtaposition of the two open source frameworks, but it looks like neither option overwhelmingly comes out on top. Writer Sreejani Bhattacharyya begins:
“As companies are deploying more and more machine learning models into their systems, a variety of frameworks (some open-source, some not) have come up over the years to make this deployment faster and more efficient. Some of the popular frameworks include TensorFlow, Amazon SageMaker, IBM Watson Studio, Google Cloud AutoML, and Azure Machine Learning Studio, among others. Tensorflow, by far, takes one of the top spots when it comes to machine learning frameworks that technologists depend on. Recently, PyCaret, a low-code machine learning library in Python, has also become increasingly popular among ML practitioners. Let us take a look at how both of them work and what makes them different from each other.”
Bhattacharyya reminds us Google-developed TensorFlow recently celebrated its sixth birthday. At first it was designed for internal Googler use, but was later released under the Apache License, Version 2.0. Since then, a wealth of tools and libraries have grown up around the framework. TensorFlow works with several programming languages, including Python, C++, JavaScript, and Java and can be run on multiple CPUs and GPUs running macOS, Windows, Android, iOS, or 64-bit Linux. The most recent version 2.7.0 improved several features, like debugging, public convolution, and data service auto-sharding.
So why even consider a newcomer like PyCaret, which launched just last November? Code efficiency. We learn:
“Also open source in nature, PyCaret is a low-code machine learning library in Python. It helps data scientists perform end-to-end experiments efficiently. It allows them to move from preparing data to deploying their model within minutes. … Pycaret is rising in popularity in comparison to other ML libraries, as it provides an alternate low-code library that can perform complex machine learning tasks with only a few lines of code. It is built around several machine learning libraries and frameworks such as scikit-learn, XGBoost, Microsoft LightGBM, and spaCy, among others.”
The write-up goes on to describe the working areas of each framework and concludes with a summary of the advantages of each. In a nutshell, TensorFlow excels in performance, scalability, and library management. PyCaret has the edge in productivity and being business-ready. Bhattacharyya found both easy to deploy. Both of these ML frameworks are good options—the choice really comes down to one’s needs and preferences.
Cynthia Murrell, February 2, 2022
Opinion Shaping Shifts into High Gear
January 25, 2022
Two interesting opinion shaping initiatives caught my attention. Both are based on digital constructions but will make use of fuddy duddy print and television as warranted. Let’s look at each quickly and then step back and figure out what’s the context of the two activities.
The first is explained in two write ups. The first is “The UK Government Is Reportedly Preparing a PR Blitz against End-to-End Encryption.” The idea is that making it more difficult to access a British citizen’s messages is not a good idea. There are nuances, of course. Certain government agencies want ASCII, easy access, and real time content pipes. A related story is “UK Government Readies Anti-Encryption Publicity Campaign to ‘Keep Children Safe’ Online.” One cannot argue with the keep children safe angle. Who doesn’t want that? In order to achieve “safe,” it is necessary to have access to user content; for example, actors messages sent via Threema, the Swiss messaging system.
The second initiative is explained in “Big Tech Foes Launch Campaign Style Initiative to Push for Antitrust Legislation.” The idea here is that some wealthy people are concerned about big tech. Did big tech make these concerned individuals wealthy? Well, that’s another issue. The push is designed to build support for clipping the wings of outfits which are not doing what the big tech foes find acceptable. The idea is to use the Fancy Dan methods of political campaigns, ad agencies, and big tech to get this antitrust breakup thing done.
Here’s my take on these two initiatives:
- Both are very upfront propaganda initiatives. Each is designed to result in changes to technology. Technology has been the lead dog for too long. Now the humans are going to use technology to put the lead dogs in the kennel.
- The pivot for each is the elimination of negative capabilities like social media or encryption. Tech is bad; change is needed.
- The initiatives are likely to further fractionalize discussions of the issues sucked into these quite visible programs. There’s nothing like starting a discussion with one side asking, “Are you in favor of child abuse?”
Net net: These are externalizing activities and make clear that methods once kept under wraps are now on public display. Good or bad? It depends on how one answers those tough questions?
Stephen E Arnold, January 25, 2022
A Comparison: US Vs. European Government Methods
January 21, 2022
I know one thing about 5G. The T Mobile super high speed service delivers data more slowly than my 4G / LTE service. Thus, it is difficult for me to accept that the pig slow 5G in rural Kentucky is a threat to aircraft eager to land on the dirt road used by certain characters in the Commonwealth.
I noted “5G Is Grounding Planes and Freaking Out Airlines: We Found Out Why.” I want to sidestep the somewhat interesting discussion about who shot John, the 5G expert. The US government and the airlines are wrestling with US 5G carriers. The main idea is a minor one; that is, 5G signals in the C band emitted from vertically mounted towers could — note the word could — cause an aircraft to demonstrate one of Newton’s Laws in an expensive way.
But here’s the quote which caught my attention:
The issues haven’t affected other countries as badly because they don’t use the same 5G frequencies as the US. In Europe, for instance, the network operates on a wavelength that is less likely to cause interference. Both the EU’s Aviation Safety Authority and the UK’s Civil Aviation Authority say there’s no such problem with their networks. China and Australia have also rolled out 5G without any issues with aircraft…. Critics have also pointed the finger at the federal government. They’ve blamed the Trump administration for failing to create a national spectrum policy and the Biden administration for the chaotic rollout. Somehow, Europe’s collection of crappy governments has avoided such problems. [Emphasis added by the Beyond Search editor]
Interesting. Now European governments have a larger challenge to surmount. Vacationing in Kiyv perhaps?
Stephen E Arnold, January 21, 2022
HP: A Minor Glitch from the Outfit Purchasing Autonomy
December 30, 2021
My Japanese is non existent, so I rely on Google Translate. (The Beyond Search team loves the Google.) According to the Institute for Information Management and Communication at Kyoto University, an HP supercomputer system experience a momentary lapse. (Possibly the same type of issue which sparked the purchase of Autonomy?) Google Translate offers this output:
An accident occurred in which some data in the capacity storage (/ LARGE0) was unintentionally deleted.
How much data were lost? Not much. Just 77 terabytes or the equivalent of 77 trillion bytes. This worked out to 34 million files affecting 14 groups of researchers. Good news some back ups were available but for four of those groups, the data are now gone, or in the phraseology of social media, cancelled.
The apology appears in red, which makes the problem go away.
Who beavered away on the system? Nippon Hewlett Packard.
Wasn’t HP the outfit allegedly paying very close attention to its board of directors’ behavior? Maybe not. But HP definitely bought Autonomy and HP definitely will be unpopular with the four groups losing their data.
Happy New Year. And what’s the New Year’s resolution for the Institute for Information Management and Communication? Buy more HP is one possibility.
Stephen E Arnold, December 31, 2021
Lick Tech: A Taste for Covid and Leukoplakia?
December 30, 2021
I spotted a remarkable photo in this online story: “Lick It Up: Japan Prof Creates ‘Tele-Taste’ TV Screen.” Here’s the main point of the write up:
Japanese professor has developed a prototype lickable TV screen that can imitate food flavors, another step toward creating a multisensory viewing experience. The device, called Taste the TV (TTTV), uses a carousel of 10 flavor canisters that spray in combination to create the taste of a particular food. The flavor sample then rolls on hygienic film over a flat TV screen for the viewer to try.
I am reluctant to replicate the image in the online publication. Visualize this. One approaches a large TV screen and licks the panel as one would an ice cream cone on the beach in Guarujá.
Think of this approach:
Winner!
Stephen E Arnold, December 30, 2021
The CEOs Have a New Best Friend: Information Technology People
December 21, 2021
During the COVID-19 pandemic, organizational leaders turned to their IT departments to keep business running. IT professionals were propelled into leadership roles and were integral to keeping the entire globe from crumbling. ZD Net explains how, “Technology Leaders’ Influence In Their Businesses Grows Beyond Expectations.”
Snow Software recently conducted a survey of 1000 IT leaders and discovered that they faired well during the world crisis. Their roles became so important that they will venture into leadership positions that fall outside IT. Before COVID-19, 89% of respondents said that their roles were undervalued. Now 90% claim they are viewed as trusted advisors within their companies. Ninety-four percent of IT leaders view innovation as an integral part of their jobs, but 71% said they are spending too much time fixing problems than the former.
In 2022, IT leaders want to concentrate on reducing IT costs, improving customer relations, and improving day-to-ay operations. The survey showed:
“‘These areas of focus can often conflict, or at least compete, with one another which, so often, is an all-too-familiar pain felt by IT leaders,’ the researchers observe. ‘To balance these priorities moving forward, CIOs need a more advanced approach for managing their technology environment. Nearly all respondents say they are in the process of adopting the cloud in some form — and 61% increased their use of cloud services over the last year.”
It is great that IT leaders are influencing business decisions and shaping the work environment, but will things revert back as the world heads towards normalcy? Also since IT is a white, male dominated industry will this exacerbate the bro culture that dominates industry?
Sounds like a winner for the techies who happen to be male.
Whitney Grace, December 21, 2021
Scientific Research Might Not Work The Second Time Around or the First Time Either
December 20, 2021
Scientific research is one way humanity advances, but Science Alert brings into question if studies’ results can be replicated: “Strenuous 8-Year Effort To Replicate Key Cancer Research Finds An Unwelcome Surprise.” Common sense and the scientific process tells that if results cannot be replicated a second time, then they are not going to work. Cancer research is facings a stigma about scientific studies being replicated:
“The research looked at 193 different experiments found in 53 cancer-related papers published in high-profile journals between 2010 and 2012, and found that none of the experiments could be set up again using only the information published. After getting help from the original study authors, 50 experiments from 23 papers were reproduced.
That only a quarter of the experiments could be rerun at all is concerning – some of the original authors never responded to requests for help – but the results showed that these reproduced tests showed effect sizes that were often smaller than what the original studies yielded.”
The findings of the replicated studies discovered that the evidence was weaker than the original experiments. This does not mean that findings are false, but further testing is needed. Furthermore, time, money, and resources are wasted in clinical trials on patients where drugs do not affect diseases. Demands for results shape cancer biology and other scientific research.
These mounting pressures hinder scientific research and delay eventual cancer cures. There is a saying, “Art for art’s sake,” so why cannot there not be “Research for research’s sake” in order to advance science? Plus one can make up data, fiddle the results, or contact colleagues for some STM SEO goodness.
Whitney Grace, December 20, 2021
Quantum Supremacy Questioned
November 25, 2021
IBM is the quantum supremacist. Google was the previous quantum cage match PR champ. What’s up with quantum supremacy other than buzzwords, public relations hoo hah, and worry lines that encryption will die?
An interesting take on the Google quantum thing appears in “Math May Have Caught Up with Google’s Quantum-Supremacy Claims.” The article is a gilding of a tidy green sward with a couple of Swiss Fleckvieh contributions steaming in the morning sun.
The write up reports:
Google chose a very specific method of computing the expected behavior of its processor, but there are other ways of doing equivalent computations. Over the intervening time, a few options have been explored that do perform better. Now, Feng Pan, Keyang Chen, and Pan Zhang are describing a specific method that allows a GPU-based cluster to produce an equivalent output in only 15 hours. Run it on a leading supercomputer, and they estimate that it would outperform the Sycamore quantum processor.
Parse this and then summarize: Google pulled a high school science club method from its hip pocket.
I also noted this statement in the write up:
In our chat with Darío Gil, head of IBM research, he dismissed the idea of quantum supremacy and instead focused on getting to what he termed quantum advantage: where quantum computers consistently outperform classical ones on problems that are useful for companies. So, unless someone else wants to pay IBM to reserve the time needed to perform Google’s computations on IBM’s hardware, this is likely to get fairly academic.
One tiny problem: IBM seems to imply that it’s the big dog in quantum computing if I understand the information in “First Quantum Computer to Pack 100 Qubits Enters Crowded Race.”
Yep, got it.
Stephen E Arnold, November 25, 2021
Psychopathy: Do the Patients Referenced by Richard Kraft Ebing Gravitate to Work in High Tech?
November 12, 2021
First, who is Richard Kraft Ebing? He was an Austro-German psychiatrist with some interesting research. Wowza. He described selected human behaviors in a way which caught the attention of a couple of the Psychology Today professional when we were talking after I delivered a report. Yep, that was a memorable day. The big dog in overalls; the marketing wizard chatting intensely with an intern in gym clothes; and the sun sparkling on the beach behind the house in Del Mar, California. I recall there was some talk about the computer company providing hardware and software to the firm which owned Psychology Today, Intellectual Digest, and a few other high IQ publications. The main point was that the computer sales people lied. “Those guys cheated us. We were raped.” That’s when I referenced good old Richard Kraft Ebing?
Flash forward to “Science Reveals the Fascinating Link between Lying and Technology.” The story is paywalled, of course. One pays for the truth, Silicon Valley infused journalism, and the unvarnished truth about high technology in its assorted manifestation.
But before looking that the article itself, let me highlight two of the rules for high technology sales and and marketing effectuators.
Rule Number One; herewith:
Tell the prospect what he or she wants to hear.
Now for Rule Number Two:
Hyperbole and vaporware are not really falsehoods. Sell sizzle, not steak.
The article in Fast Company is quite like some of T George Harris’ faves. (T George, described as a visionary journalist, was a big wheel at the outfit which owned Psych Today and ID decades ago.)
The main point of the write up published online on November 12, 2021, struck me as:
The belief that lying is rampant in the digital age just doesn’t match the data.
There you go. Definitive evidence that truth reigns supreme. Example: when Verizon uses the word “unlimited.” Example: Charter Spectrum sells 200 megabit connectivity. Example: FAANG statements under oath.
Yep, truth, integrity, and the best of what’s good for “users.” Psychopathia whatever.
Stephen E Arnold, November 12, 2021
China, Smart Software, and Different Opinions
October 21, 2021
I spotted “China Isn’t the AI Juggernaut the West Fears.” The main idea for the story is that China has cornered smart software applications and innovation. Therefore, the future — at least some of it — is firmly in the grip of the Chinese Communist Party.
My hunch is that this article in the Japan Times is a response to articles like “Former Senior Pentagon Official Says China is Kicking Our Ass in Artificial Intelligence.” Nicolas Chaillan, a former Pentagon official, suggested that China is making significant progress in AI. If China continues on its present path, that country may surpass the US and its allies in smart software.
What’s interesting is that quite different viewpoints are zooming around the interwebs.
The Japan Times’ take which channels Bloomberg includes this statement:
On paper, the U.S. and China appear neck and neck in artificial intelligence. China leads in the share of journal citations — helped by the fact that it also publishes more — while the U.S. is far ahead in the more qualitative metric of cited conference papers, according to a recent report compiled by Stanford University. So while the world’s most populous country is an AI superpower, investors and China watchers shouldn’t put too much stock in the notion that its position is unassailable or that the U.S. is weaker. By miscalculating the others’ abilities, both superpowers risk overestimating their adversary’s strengths and overcompensating in a way that could lead to a Cold War-style AI arms race.
Yep, citation analysis.
I don’t have a dog in this fight. I want to point out that citation analysis, like patent documents, may not tell a comprehensive story.
I would suggest that citation analysis may be distorted by the search engine optimization techniques used by some academics and government-funded researchers. In addition, the publication flow from what I call AI cabals — loose federations of like minded researchers who cross cite one another — provide a fun house mirror opportunity.
That is, what’s reflected is a version of reality, not the reality that a person like myself would perceive without the mirrors.
Net net: The Japan Times’ write up may be off the mark. As a result, the view point of Nicolas Chaillan may warrant serious consideration.
Stephen E Arnold, October 21, 2021
-
- Subscribe to Beyond Search
Feature archive
News archive
-
