Small Data Also Crucial, With Tested Transfer Assumption

We tried related ideas but found that the transfer assumption rarely held well enough to be useful.  Useful to keep in mind to potentially simplify models, with care..

'Small Data' Is Also Crucial for Machine Learning

By Scientific American, October 19, 2021

The existence of techniques such as transfer learning does not seem to have reached the awareness of policy makers and business leaders making decisions about AI funding and adoption..... 

Some of the most prominent artificial intelligence (AI) breakthroughs in the past decade have relied on enormous data sets. But AI is not only about large data sets; research in "small data" approaches has grown extensively over the past decade—with so-called transfer learning as an especially promising example.

Small-data approaches such as transfer learning offer numerous advantages over more data-intensive methods. Enabling the use of AI with less data can bolster progress in areas where little or no data exists, such as in forecasting natural hazards that occur relatively rarely or in predicting the risk of disease for a population set that does not have digital health records.

From Scientific American

View Full Article

Netflix is Shelf Selling with AI

 When I first saw the long rows of 'content' on a Netflix selection shelf.   I immediately thought of a supermarket shelf,  which I had much experience with.   This was not alphabetic or random.  And certainly not stationary.  Each time I went back things had changed.    But how?  They were selling to me with a purpose.  Ratings?   Increasing viewers.  With AI.   Part of the idea below; 

Netflix explains how it uses AI to sell you on a show

Machines may have a better sense of what will draw you in.  In Engadget By Jon Fingas, @jonfingas

If Netflix’s decisions on marketing its hundreds of original shows seem highly calculated... that’s probably because they were. Netflix has outlined how it uses AI to market shows and predict their success in ways that conventional box office numbers and Nielsen ratings likely couldn’t match. Effectively, it comes down to finding connections and determining the likely audience sizes.

The method relies on transfer learning, where the the parameters learned from a “source task” improve the performance of a “target task.” In this case, the source tasks are simple: what titles are comparable to a Netflix original, and what kind of viewership can the service expect?

For thematic comparisons, Netflix creates a “similarity map” where AI uses a show’s metadata, tags and summaries (“embeddings” in Netflix’s world) help determine links to other titles. Marketers would know which shows and movies to help describe a coming-of-age comedy, for example.

With audience sizes, the service has an AI model that compares the audience sizes of similar work in a given country. If a drama is likely to fare well in Spain, Netflix might not only ramp up marketing in the region but prepare dubs and subtitles earlier.

The systems are self-supervised, letting them access a much wider range of titles than they would if they were limited to Netflix’s own info.   ... "

Causality is Important for Machine Learning

Have always thought causal thinking and learning was a major consideration for the future of AI in general.  Here a step in the right direction.  But overall its still a hard question.

Understanding Causality Is the Next Challenge for Machine Learning

Teaching robots to understand "why" could help them transfer their knowledge to other environments  By Payal Dhar

“Causality is very important for the next steps of progress of machine learning,” said Yoshua Bengio, a Turing Award-wining scientist known for his work in deep learning, in an interview with IEEE Spectrum in 2019. So far, deep learning has comprised learning from static datasets, which makes AI really good at tasks related to correlations and associations. However, neural nets do not interpret cause-and effect, or why these associations and correlations exist. Nor are they particularly good at tasks that involve imagination, reasoning, and planning. This, in turn, limits AI from being able to generalize their learning and transfer their skills to another related environment.

The lack of generalization is a big problem, says Ossama Ahmed, a master’s student at ETH Zurich who has worked with Bengio’s team to develop a robotic benchmarking tool for causality and transfer learning. “Robots are [often] trained in simulation, and then when you try to deploy [them] in the real world…they usually fail to transfer their learned skills. One of the reasons is that the physical properties of the simulation are quite different from the real world,” says Ahmed. The group’s tool, called CausalWorld, demonstrates that with some of the methods currently available, the generalization capabilities of robots aren’t good enough—at least not to the extent that “we can deploy [them] safely in any arbitrary situation in the real world,” says Ahmed.

The paper on CausalWorld   , available as a preprint, describes benchmarks in a simulated robotics manipulation environment using the open-source TriFinger robotics platform. The main purpose of CausalWorld is to accelerate research in causal structure and transfer learning using this simulated environment, where learned skills could potentially be transferred to the real world. Robotic agents can be given tasks that comprise pushing, stacking, placing, and so on, informed by how children have been observed to play with blocks and learn to build complex structures. There is a large set of parameters, such as weight, shape, and appearance of the blocks and the robot itself, on which the user can intervene at any point to evaluate the robot’s generalization capabilities. ...  '

A Look at Transfer Learning

Good generalized look at the concept of Transfer Learnig

Everything you need to know about transfer learning in AI   in TNW

Today, artificial intelligence programs can recognize faces and objects in photos and videos, transcribe audio in real-time, detect cancer in x-ray scans years in advance, and compete with humans in some of the most complicated games.

Until a few years ago, all these challenges were either thought insurmountable, decades away, or were being solved with sub-optimal results. But advances in neural networks and deep learning, a branch of AI that has become very popular in the past few years, has helped computers solve these and many other complicated problems.

Unfortunately, when created from scratch, deep learning models require access to vast amounts of data and compute resources. This is a luxury that many can’t afford. Moreover, it takes a long time to train deep learning models to perform tasks, which is not suitable for use cases that have a short time budget.

Fortunately, transfer learning, the discipline of using the knowledge gained from one trained AI model to another, can help solve these problems.

The cost of training deep learning models
Deep learning is a subset of machine learning, the science of developing AI through training examples. The concepts and science behind deep learning and neural networks is as old as the term “artificial intelligence” itself. But until recent years, they had been largely dismissed by the AI community for being inefficient.

The availability of vast amounts of data and compute resources in the past few years have pushed neural networks into the limelight and made it possible to develop deep learning algorithms that can solve real world problems.

To train a deep learning model, you basically must feed a neural network with lots of annotated examples. These examples can be things such as labeled images of objects or mammograms scans of patients with their eventual outcomes. The neural network will carefully analyze and compare the images and develop mathematical models that represent the recurring patterns between images of a similar category.

[Read: Weird AI illustrates why algorithms still need people]

There already exists several large open-source datasets such as ImageNet, a database of more than 14 million images labeled in 22,000 categories, and MNIST, a dataset of 60,000 handwritten digits. AI engineers can use these sources to train their deep learning models.

However, training deep learning models also requires access to very strong computing resources. Developers usually use clusters of CPUs, GPUs or specialized hardware such as Google’s Tensor Processors (TPUs) to train neural networks in a time-efficient way. The costs of purchasing or renting such resources can be beyond the budget of individual developers or small organizations. Also, for many problems, there aren’t enough examples to train robust AI models.

Transfer learning makes deep learning training much less demanding
Say an AI engineer wants to create an image classifier neural network to solve a specific problem. Instead of gathering thousands and millions of images, the engineer can use one of the publicly available datasets such as ImageNet and enhance it with domain-specific photos.

But the AI engineer must still rent pay a hefty sum to rent the compute resources necessary to run those millions of images through the neural network. This is where transfer learning comes into play. Transfer learning is the process of creating new AI models by fine-tuning previously trained neural networks.  ... " 

Simulating Molecular Motion with Neural Networks

Have now seen several examples of using neural nets to simulate complex systems by generating a model from data.    Thought provoking example I am following up on.

Researchers Cast Neural Nets to Simulate Molecular Motion 
Los Alamos National Laboratory News
By Nancy Ambrosiano      July 2, 2019

The U.S. Department of Energy's Los Alamos National Laboratory (LANL), the University of North Carolina at Chapel Hill, and the University of Florida demonstrated that artificial neural nets can be taught to encode quantum mechanical laws that define molecular motion, potentially advancing simulations across many disciplines. Said LANL's Justin Smith, "We can now model materials and molecular dynamics billions of times faster compared to conventional quantum methods, while retaining the same level of accuracy." The researchers developed a machine learning technique to build empirical potentials—atomic dynamics descriptions that follow classical physical and Newtonian laws—from data collected about millions of compounds. The transfer learning technique can be applied to new molecules in milliseconds .... " 

Researchers cast neural nets to simulate molecular motion
Machine learning allows quantum mechanics to be efficiently applied to molecular simulations for drug development, detonation physics and more .... " 


Publication:   J. S. Smith, B. T. Nebgen, R. Zubatyuk, N. Lubbers, C. Devereux, K. Barros, S. Tretiak, O. Isayev, A. E. Roitberg, “Approaching coupled cluster accuracy with a general-purpose neural network potential through transfer learning,” Nature Communications 10.1038/s41467-019-10827-4 (2019)  ..... " 

Technical paper:  https://www.nature.com/articles/s41467-019-10827-4

On Transfer Learning

Short, non technical introduction,with cautions.

Transfer learning: the dos and don’ts      By Chris von Csefalvay from Starchema Blog

If you have recently started doing work in deep learning, especially image recognition, you might have seen the abundance of blog posts all over the internet, promising to teach you how to build a world-class image classifier in a dozen or fewer lines and just a few minutes on a modern GPU. What’s shocking is not the promise but the fact that most of these tutorials end up delivering on it. How is that possible? To those trained in ‘conventional’ machine learning techniques, the very idea that a model developed for one data set could simply be applied to a different one sounds absurd.

The answer is, of course, transfer learning, one of the most fascinating features of deep neural networks. In this post, we’ll first look at what transfer learning is, when it will work, when it might work, and why it won’t work in some cases, finally concluding with some pointers at best practices for transfer learning. ...."

See also: https://en.wikipedia.org/wiki/Transfer_learning

Definition and Use of Transfer Learning

Was pointed out to me that transfer learning was important to leveraging intelligence.    I had heard of it, but sought out a closer look at definition and areas of useful research, here is a start.  Not mentioned in any of the usual Quant methods in business books, but even starting with smaller or restricted prototypes can be seen as transfer learning.

Understanding Transfer Learning ....

Transfer learning,   From Wikipedia, the free encyclopedia: 
Transfer learning or inductive transfer is a research problem in machine learning that focuses on storing knowledge gained while solving one problem and applying it to a different but related problem.[1] For example, knowledge gained while learning to recognize cars could apply when trying to recognize trucks. This area of research bears some relation to the long history of psychological literature on transfer of learning, although formal ties between the two fields are limited.  .... " 

" ... The earliest cited work on transfer in machine learning is attributed to Lorien Pratt, who formulated the discriminability-based transfer (DBT) algorithm in 1993.[2] ... In 1997, the journal Machine Learning published a special issue devoted to transfer learning,[3] and by 1998, the field had advanced to include multi-task learning,[4] along with a more formal analysis of its theoretical foundations.[5] Learning to Learn,[6] edited by Pratt and Sebastian Thrun, is a 1998 review of the subject. .... Transfer learning has also been applied in cognitive science, with the journal Connection Science publishing a special issue on reuse of neural networks through transfer in 1996 ... " 

A Survey on Transfer Learning:

ACM article abstract.

A major assumption in many machine learning and data mining algorithms is that the training and future data must be in the same feature space and have the same distribution. However, in many real-world applications, this assumption may not hold. For example, we sometimes have a classification task in one domain of interest, but we only have sufficient training data in another domain of interest, where the latter data may be in a different feature space or follow a different data distribution. In such cases, knowledge transfer, if done successfully, would greatly improve the performance of learning by avoiding much expensive data-labeling efforts. 

In recent years, transfer learning has emerged as a new learning framework to address this problem. This survey focuses on categorizing and reviewing the current progress on transfer learning for classification, regression, and clustering problems. In this survey, we discuss the relationship between transfer learning and other related machine learning techniques such as domain adaptation, multitask learning and sample selection bias, as well as covariate shift. We also explore some potential future issues in transfer learning research ... " 

Areas of AI to Watch

Very good piece by Nathan Benaich,  Covers at the right Tech savvy management-introductory level a number of AI oriented techniques.  I like these because each is more problem than technology facing.  Also brings in practical decision approaches we used in operations research.   Sections on each of these, have used a number of these, though sometimes under different names.   Note the less data article, critical.

Reinforcement Learning  
Generative Models
Networks with Memory
Learning from Less Data  and Building Small models  (Transfer Learning) 
Hardware for Training and Inference
Simulation Environments

Mentions of many companies and researchers involved in each area,  largely non-technical descriptions of each area.   Good read for all with strong interest in this area.

Nathan has " ...  an analytical digest of artificial intelligence and machine learning news from the technology industry, research lab and venture capital market  ...  ".     Link to it in the article,  I have subscribed.

Transfer Learning for AI Projects

Had always thought that intelligence was about learning, so this concept struck me.   Note mention of improbable events and model correctness maintenance,  always of concern in such studies.  Technical.

'Transfer learning' jump-starts new AI projects
Machine learning, once implemented, tends to be specific to the data and requirements of the task at hand. Transfer learning is the act of abstracting and reusing those smarts

'Transfer Learning' Jump-Starts New AI Projects  in InfoWorld by James Kobielus

Abstracting and reusing knowledge gleaned from a machine-learning application in other, newer apps--or "transfer learning"--is supplementing other learning methods that constitute the backbone of most data science practices. Among the technique's practical uses is productivity acceleration modeling, which is viable when prior work can be reused without extensive revision in order to speed up time to insight. Another transfer-learning application involves the method helping scientists produce machine-learning models that exploit relevant training data from prior modeling projects.

This technique is particularly appropriate for addressing projects in which prior training data can easily become obsolete, which is a problem that frequently occurs in dynamic problem domains. A third area of data science in which transfer learning could yield benefits is risk mitigation. In this situation, transfer learning can help scientists leverage subsets of training data and feature models from related domains when the underlying conditions of the modeled phenomenon have radically changed. 

This can help researchers ameliorate the risk of machine-learning-driven predictions in any problem domain vulnerable to extremely improbable events. Transfer learning also is critical to data scientists' efforts to create "master learning algorithms" that automatically obtain and apply fresh contextual knowledge via deep neural networks and other forms of artificial intelligence. ... "