Connect with us

AI

Why machine learning struggles with causality

Avatar

Published

on

Join Transform 2021 for the most important themes in enterprise AI & Data. Learn more.


When you look at a baseball player hitting the ball, you can make inferences about causal relations between different elements. For instance, you can see the bat and the baseball player’s arm moving in unison, but you also know that it is the player’s arm that is causing the bat’s movement and not the other way around. You also don’t need to be told that the bat is causing the sudden change in the ball’s direction.

Likewise, you can think about counterfactuals, such as what would happen if the ball flew a bit higher and didn’t hit the bat.

Baseball batter hitting the ball

Such inferences come to us humans intuitively. We learn them at a very early age, without being explicitly instructed by anyone and just by observing the world. But for machine learning algorithms, which have managed to outperform humans in complicated tasks such as go and chess, causality remains a challenge. Machine learning algorithms, especially deep neural networks, are especially good at ferreting out subtle patterns in huge sets of data. They can transcribe audio in real-time, label thousands of images and video frames per second, and examine x-ray and MRI scans for cancerous patterns. But they struggle to make simple causal inferences like the ones we just saw in the baseball video above.

In a paper titled “Towards Causal Representation Learning,” researchers at the Max Planck Institute for Intelligent Systems, the Montreal Institute for Learning Algorithms (Mila), and Google Research, discuss the challenges arising from the lack of causal representations in machine learning models and provide directions for creating artificial intelligence systems that can learn causal representations.

This is one of several efforts that aim to explore and solve machine learning’s lack of causality, which can be key to overcoming some of the major challenges the field faces today.

Independent and identically distributed data

Why do machine learning models fail at generalizing beyond their narrow domains and training data?

“Machine learning often disregards information that animals use heavily: interventions in the world, domain shifts, temporal structure — by and large, we consider these factors a nuisance and try to engineer them away,” write the authors of the causal representation learning paper. “In accordance with this, the majority of current successes of machine learning boil down to large scale pattern recognition on suitably collected independent and identically distributed (i.i.d.) data.”

i.i.d. is a term often used in machine learning. It supposes that random observations in a problem space are not dependent on each other and have a constant probability of occurring. The simplest example of i.i.d. is flipping a coin or tossing a die. The result of each new flip or toss is independent of previous ones and the probability of each outcome remains constant.

When it comes to more complicated areas such as computer vision, machine learning engineers try to turn the problem into an i.i.d. domain by training the model on very large corpora of examples. The assumption is that, with enough examples, the machine learning model will be able to encode the general distribution of the problem into its parameters. But in the real world, distributions often change due to factors that cannot be considered and controlled in the training data. For instance, convolutional neural networks trained on millions of images can fail when they see objects under new lighting conditions or from slightly different angles or against new backgrounds.

Training datasets and real world objects

Above: Objects in training datasets vs objects in the real world (source: objectnet.dev)

Image Credit: TechTalks

Efforts to address these problems mostly include training machine learning models on more examples. But as the environment grows in complexity, it becomes impossible to cover the entire distribution by adding more training examples. This is especially true in domains where AI agents must interact with the world, such as robotics and self-driving cars. Lack of causal understanding makes it very hard to make predictions and deal with novel situations. This is why you see self-driving cars make weird and dangerous mistakes even after having trained for millions of miles.

“Generalizing well outside the i.i.d. setting requires learning not mere statistical associations between variables, but an underlying causal model,” the AI researchers write.

Causal models also allow humans to repurpose previously gained knowledge for new domains. For instance, when you learn a real-time strategy game such as Warcraft, you can quickly apply your knowledge to other similar games StarCraft and Age of Empires. Transfer learning in machine learning algorithms, however, is limited to very superficial uses, such as finetuning an image classifier to detect new types of objects. In more complex tasks, such as learning video games, machine learning models need huge amounts of training (thousands of years’ worth of play) and respond poorly to minor changes in the environment (e.g., playing on a new map or with a slight change to the rules).

“When learning a causal model, one should thus require fewer examples to adapt as most knowledge, i.e., modules, can be reused without further training,” the authors of the causal machine learning paper write.

Causal learning

So, why has i.i.d. remained the dominant form of machine learning despite its known weaknesses? Pure observation-based approaches are scalable. You can continue to achieve incremental gains in accuracy by adding more training data, and you can speed up the training process by adding more compute power. In fact, one of the key factors behind the recent success of deep learning is the availability of more data and stronger processors.

i.i.d.-based models are also easy to evaluate: Take a large dataset, split it into training and test sets, tune the model on the training data, and validate its performance by measuring the accuracy of its predictions on the test set. Continue the training until you reach the accuracy you require. There are already many public datasets that provide such benchmarks, such as ImageNet, CIFAR-10, and MNIST. There are also task-specific datasets such as the COVIDx dataset for covid-19 diagnosis and the Wisconsin Breast Cancer Diagnosis dataset. In all cases, the challenge is the same: Develop a machine learning model that can predict outcomes based on statistical regularities.

But as the AI researchers observe in their paper, accurate predictions are often not sufficient to inform decision-making. For instance, during the coronavirus pandemic, many machine learning systems began to fail because they had been trained on statistical regularities instead of causal relations. As life patterns changed, the accuracy of the models dropped.

Causal models remain robust when interventions change the statistical distributions of a problem. For instance, when you see an object for the first time, your mind will subconsciously factor out lighting from its appearance. That’s why, in general, you can recognize the object when you see it under new lighting conditions.

Causal models also allow us to respond to situations we haven’t seen before and think about counterfactuals. We don’t need to drive a car off a cliff to know what will happen. Counterfactuals play an important role in cutting down the number of training examples a machine learning model needs.

Causality can also be crucial to dealing with adversarial attacks, subtle manipulations that force machine learning systems to fail in unexpected ways. “These attacks clearly constitute violations of the i.i.d. assumption that underlies statistical machine learning,” the authors of the paper write, adding that adversarial vulnerabilities are proof of the differences in the robustness mechanisms of human intelligence and machine learning algorithms. The researchers also suggest that causality can be a possible defense against adversarial attacks.

Above: Adversarial attacks target machine learning’s sensitivity to i.i.d. In this image, adding a imperceptible layer of noise to this panda picture causes a convolutional neural network to mistake it for a gibbon.

Image Credit: TechTalks

In a broad sense, causality can address machine learning’s lack of generalization. “It is fair to say that much of the current practice (of solving i.i.d. benchmark problems) and most theoretical results (about generalization in i.i.d. settings) fail to tackle the hard open challenge of generalization across problems,” the researchers write.

Adding causality to machine learning

In their paper, the AI researchers bring together several concepts and principles that can be essential to creating causal machine learning models.

Two of these concepts include “structural causal models” and “independent causal mechanisms.” In general, the principles state that instead of looking for superficial statistical correlations, an AI system should be able to identify causal variables and separate their effects on the environment.

This is the mechanism that enables you to detect different objects regardless of the view angle, background, lighting, and other noise. Disentangling these causal variables will make AI systems more robust against unpredictable changes and interventions. As a result, causal AI models won’t need huge training datasets.

“Once a causal model is available, either by external human knowledge or a learning process, causal reasoning allows to draw conclusions on the effect of interventions, counterfactuals and potential outcomes,” the authors of the causal machine learning paper write.

The authors also explore how these concepts can be applied to different branches of machine learning, including reinforcement learning, which is crucial to problems where an intelligent agent relies a lot on exploring environments and discovering solutions through trial and error. Causal structures can help make the training of reinforcement learning more efficient by allowing them to make informed decisions from the start of their training instead of taking random and irrational actions.

The researchers provide ideas for AI systems that combine machine learning mechanisms and structural causal models: “To combine structural causal modeling and representation learning, we should strive to embed an SCM into larger machine learning models whose inputs and outputs may be high-dimensional and unstructured, but whose inner workings are at least partly governed by an SCM (that can be parameterized with a neural network). The result may be a modular architecture, where the different modules can be individually fine-tuned and re-purposed for new tasks.”

Such concepts bring us closer to the modular approach the human mind uses (at least as far as we know) to link and reuse knowledge and skills across different domains and areas of the brain.

Above: Combining causal graphs with machine learning will enable AI agents to create modules that can be applied to different tasks without much training

Combining causal graphs with machine learning will enable AI agents to create modules that can be applied to different tasks without much training
It is worth noting, however, that the ideas presented in the paper are at the conceptual level. As the authors acknowledge, implementing these concepts faces several challenges: “(a) in many cases, we need to infer abstract causal variables from the available low-level input features; (b) there is no consensus on which aspects of the data reveal causal relations; (c) the usual experimental protocol of training and test set may not be sufficient for inferring and evaluating causal relations on existing data sets, and we may need to create new benchmarks, for example with access to environment information and interventions; (d) even in the limited cases we understand, we often lack scalable and numerically sound algorithms.”

But what’s interesting is that the researchers draw inspiration from much of the parallel work being done in the field. The paper contains references to the work done by Judea Pearl, a Turing Award–winning scientist best known for his work on causal inference. Pearl is a vocal critic of pure deep learning methods. Meanwhile, Yoshua Bengio, one of the co-authors of the paper and another Turing Award winner, is one of the pioneers of deep learning.

The paper also contains several ideas that overlap with the idea of hybrid AI models proposed by Gary Marcus, which combines the reasoning power of symbolic systems with the pattern recognition power of neural networks. The paper does not, however, make any direct reference to hybrid systems.

The paper is also in line with system 2 deep learning, a concept first proposed by Bengio in a talk at the NeurIPS 2019 AI conference. The idea behind system 2 deep learning is to create a type of neural network architecture that can learn higher representations from data. Higher representations are crucial to causality, reasoning, and transfer learning.

While it’s not clear which of the several proposed approaches will help solve machine learning’s causality problem, the fact that ideas from different—and often conflicting—schools of thought are coming together is guaranteed to produce interesting results.

“At its core, i.i.d. pattern recognition is but a mathematical abstraction, and causality may be essential to most forms of animate learning,” the authors write. “Until now, machine learning has neglected a full integration of causality, and this paper argues that it would indeed benefit from integrating causal concepts.”

Ben Dickson is a software engineer and the founder of TechTalks. He writes about technology, business, and politics.

This story originally appeared on Bdtechtalks.com. Copyright 2021

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact. Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://venturebeat.com/2021/03/19/why-machine-learning-struggles-with-causality/

AI

Aite survey: Financial institutions will invest more to automate loan process

Avatar

Published

on

Financial institutions plan to increase their spend on automations and collections management solutions for their loan processes. Fresh results on consumer lending practice from research and advisory firm Aite Group indicate lenders plan to invest more heavily in their collections processes, said Leslie Parrish, senior analyst for the Aite Group’s consumer lending practice. Parrish shared […]

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://bankautomationnews.com/allposts/lending/aite-survey-financial-institutions-will-invest-more-to-automate-loan-process/

Continue Reading

AI

Facial recognition, other ‘risky’ AI set for constraints in EU

Avatar

Published

on

Facial recognition and other high-risk artificial intelligence applications will face strict constraints under new rules unveiled by the European Union that threaten hefty fines for companies that don’t comply.

The European Commission, the bloc’s executive body, proposed measures on Wednesday that would ban certain AI applications in the EU, including those that exploit vulnerable groups, deploy subliminal techniques or score people’s social behavior.

The use of facial recognition and other real-time remote biometric identification systems by law enforcement would also be prohibited, unless used to prevent a terror attack, find missing children or tackle other public security emergencies.

Facial recognition is a particularly controversial form of AI. Civil liberties groups warn of the dangers of discrimination or mistaken identities when law enforcement uses the technology, which sometimes misidentifies women and people with darker skin tones. Digital rights group EDRI has warned against loopholes for public security exceptions use of the technology.

Other high-risk applications that could endanger people’s safety or legal status—such as self-driving cars, employment or asylum decisions — would have to undergo checks of their systems before deployment and face other strict obligations.

The measures are the latest attempt by the bloc to leverage the power of its vast, developed market to set global standards that companies around the world are forced to follow, much like with its General Data Protection Regulation.

The U.S. and China are home to the biggest commercial AI companies — Google and Microsoft Corp., Beijing-based Baidu, and Shenzhen-based Tencent — but if they want to sell to Europe’s consumers or businesses, they may be forced to overhaul operations.

Key Points:

  • Fines of 6% of revenue are foreseen for companies that don’t comply with bans or data requirements
  • Smaller fines are foreseen for companies that don’t comply with other requirements spelled out in the new rules
  • Legislation applies both to developers and users of high-risk AI systems
  • Providers of risky AI must subject it to a conformity assessment before deployment
  • Other obligations for high-risk AI includes use of high quality datasets, ensuring traceability of results, and human oversight to minimize risk
  • The criteria for ‘high-risk’ applications includes intended purpose, the number of potentially affected people, and the irreversibility of harm
  • AI applications with minimal risk such as AI-enabled video games or spam filters are not subject to the new rules
  • National market surveillance authorities will enforce the new rules
  • EU to establish European board of regulators to ensure harmonized enforcement of regulation across Europe
  • Rules would still need approval by the European Parliament and the bloc’s member states before becoming law, a process that can take years

—Natalia Drozdiak (Bloomberg Mercury)

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://bankautomationnews.com/allposts/comp-reg/facial-recognition-other-risky-ai-set-for-constraints-in-eu/

Continue Reading

Artificial Intelligence

Prioritizing Artificial Intelligence and Machine Learning in a Pandemic

Avatar

Published

on

AI and ML
Illustration: © IoT For All

Artificial Intelligence (AI) and Machine Learning (ML) give companies the one thing humans can’t – scalability. Over time, humans limit a businesses’ ability to scale; they can only work so many hours at a given efficiency. On the other hand, AI and ML can work around the clock with the sole focus on a given project. As organizations navigate through COVID-19’s impact and the future of a remote workforce, scalability and efficiency can be the key to an organization’s successful recovery.

Implementation Challenges

The benefits of AI and ML don’t come without their own challenges; however, the top challenges are a lack of skills and time for proper implementation. In July, Deloitte found in a survey that 69% of respondents said the skills gap for AI implementation ranged from moderate to major to extreme. Simultaneously, many companies overlook the investment it takes to build the processes and infrastructure needed for successfully training, testing, deploying, and maintaining AI and ML in their enterprise.

Such challenges often cause companies to de-prioritize AI and ML projects, especially in times of uncertainty. That has been increasingly obvious throughout the COVID-19 pandemic. But while some organizations have drawn back on their efforts, the current global state demands the greater need for AI and ML to support critical business processes. This is especially true today given the growing remote workforce, considerations for returning to the workplace and work happening in silos worldwide.

Though challenging, it is not impossible to properly implement AI and ML. In this evolving COVID-influenced business landscape, four steps are key to effectively implementing a strong AI and ML system that helps streamline critical business processes despite uncertainty and limited resources.

Identify the Problem to Be Solved

Some companies mistakenly view AI and ML projects as a ‘silver bullet’ to solve all their problems. This often results in overinflated expectations, an unfocused approach, and unsatisfactory results. Instead, companies should identify those specific problems that will have the biggest impact from implementing AI and ML solutions and be hyper-focused on solving those problems.

Select Your Data

The second step in creating a strong AI and ML algorithm is to select the source data that your algorithm will be training on. There are two main options: training on your own data or training on a larger scale data set. Based on experience, training your algorithm on your own data puts you at a disadvantage. By training on a larger scale data set, the likelihood of success increases because your data is more representative and varied. Through advanced concepts such as transfer learning, companies can use semi-trained models based on larger data sets and then train the “last mile” using their own specific content unique to their business.

Clean House

The standby rules of data management apply here – garbage in, garbage out. Ultimately, the quality and accuracy of machine learning models depend on being representative. AI and ML – fed with the right data – can streamline operations and increase the benefit of companies’ DX and cloud migration journeys.

When you’re kicking off an AI or ML project, the most critical step is to clean up the data that your algorithm will be training on, especially if you’re using your own data or models.

Make Room for Training

AI and ML are all about probability. When you ask it a question, for example, “Is this a cat?,” the results you receive are the algorithm saying, “Out of the three buckets I was trained on, the likelihood of this image being a cat is .91, the likelihood of this image being a dog is .72 and the likelihood of this image being a bird is .32.”

This is why training on varied data is so important. If your training data only includes images of cats, dogs, and birds and you ask the algorithm to analyze the picture of a crocodile, it will only respond based on the buckets it’s been trained on – cats, dogs, and birds.

If you’ve properly selected and cleaned your data, training should be an easy last step, but it’s also an opportunity to go back to the first two steps and further refine based on your training.

The front end of training an AI and ML algorithm can be time-intensive, but following these four steps can make it easier to achieve significant outcomes. Across industries, AI and ML can quickly show ROI. For example, in the insurance industry, AI and ML can help insurers quickly search contracts, so employees aren’t sifting through contracts and repositories around the globe to answer simple questions. This means time efficiencies for an industry that COVID-19 has heavily impacted.

Even better, working with a SaaS provider with experience in your industry can make this process much easier and less costly. SaaS platforms allow companies to take advantage of having all of the infrastructure, security, and pre-trained models in place to reduce the overall effort and time to value. Many platforms allow users to uptrain the predefined models with unique customer data, reducing the training effort needed for model creation. Companies can then focus on integration with their ecosystem and workflows rather than model creation itself.

Bigger Picture

Overall, businesses can soften the impact of COVID by focusing on the bigger picture with AI and ML. Implementing AI and ML projects increase business productivity despite these times of uncertainty. As we continue on the road to recovery, we need tools like AI and ML to stay focused on the bigger picture, mission-critical tasks.

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://www.iotforall.com/prioritizing-artificial-intelligence-and-machine-learning-in-a-pandemic

Continue Reading

AI

ProGlove promotes worker well-being with human digital twin technology

Avatar

Published

on

Join Transform 2021 this July 12-16. Register for the AI event of the year.


ProGlove, the company behind an ergonomic barcode scanner, has developed new tools for analyzing human processes to build a human digital twin.

“We have always been driven to have our devices narrate the story of what is really happening on the shop floor, so we added process analytics capabilities that allow for time-motion studies, visualization of the shop floor, and more,” ProGlove CEO Andreas Koenig told VentureBeat.

The company’s newest process analytics tools can complement the typical top-down perspective of applications by adding a process-as-seen view to the conventional process-as-wanted view. Most importantly, it can also provide insights that improve well-being.

Koenig said, “We are building an ecosystem that empowers the human worker to make their businesses stronger.”

ProGlove CEO Andreas Koenig

Above: ProGlove CEO Andreas Koenig

Image Credit: ProGlove

The market for barcode scanning is still going strong and is often taken for granted, given how old it is. “You have technologies like RFID that have been celebrated for being the next big thing, and yet their impact thus far hasn’t been anywhere near where most pundits expected it,” Koenig said.

Companies like Zebra, Honeywell, and Datalogic have lasted for decades by building out an ecosystem of tools to address industry needs. “What sets us apart is that we looked beyond the obvious and started with the human worker in mind,” Koenig said.

Not only is the company providing a form factor designed to meet requirements for rugged tools, this shift to analytics could further promote efficiency, quality, and ergonomics on the shop floor.

How a human digital twin works

ProGlove’s cofounders participated in Intel’s Make It Wearable Challenge, with the idea of designing a smart glove for industries. Today, ProGlove’s MARK scanner can collect six-axis motion data, including pitch, yaw, roll, and acceleration, along with timestamps, a step count, and camera data (such as barcode reading speed and the scanner ID).

Koenig’s vision goes beyond selling a product to establish the right balance between businesses’ need for profits and their obligation to ensure worker well-being. Koenig estimates that human hands deliver 70% of added value in factories and on warehouse floors. “There is no doubt that they are your most valuable resource that needs protection. Even more so since we are way more likely to experience a shortage of human workers in the warehouses across the world than having them replaced by robots, automation, or AI.”

ProGlove Insight contextualizes the collected data and lets users compare workstations and measure the workload and effort necessary to complete the tasks. Users can also visualize their shop floor, look at heatmaps, and identify best practices or efficiency blockers. After a recent smart factory lab experiment with users, DPD and Asics realized efficiency gains by as much as 20%, Koenig said.

ProGlove’s vision of the human digital twin is built on three pillars: a digital representation of onsite workers, a visualization of the shop floor, and an industrial process engineer. “The human digital twin is all about striking the right balance between businesses’ needs for profitability, efficiency, and worker well-being,” Koenig said. At the same time, it is important that the human digital twin complies with data privacy regulations and provides transparency to frontline workers around what data is being transmitted.

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact. Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://venturebeat.com/2021/04/21/proglove-promotes-worker-well-being-with-human-digital-twin-technology/

Continue Reading
Esports5 days ago

Dota 2 patch 7.29b brings nerfs to Phantom Lancer and Lifestealer amongst other hero balance changes

Esports5 days ago

Apex Legends Season 9 will add new hero, fix Banglore bugs

Esports4 days ago

Code S: Trap & Zest advance to RO8, playoff bracket set

Blockchain3 days ago

Mining Bitcoin: How to Mine Bitcoin

Esports5 days ago

New CSGO Update Makes Items Purchased From Store Non Tradable for a Week

Esports5 days ago

Radiant Valorant streamer Solista banned for cheating on live stream

Esports5 days ago

How to Calculate Steam Market Tax on CSGO Items

Fintech4 days ago

Fintech offers brokers better commissions after BID

Blockchain4 days ago

Stanislovas Tomas im Interview: „NFTs können unsere Gesellschaft verändern“

Esports4 days ago

OWL 2021 Power Rankings – #9 Guangzhou Charge

Esports4 days ago

xQc Banned From NoPixel GTA RP Server Once Again

Esports5 days ago

OWL 2021 Power Rankings – #10 Washington Justice

Esports4 days ago

CDL Challengers Elite Stage 3 Preview

PR Newswire3 days ago

Hello Pal Signs Definitive Purchase Agreement to Acquire Interest in Dogecoin/Litecoin Mining Assets

Coinbase hourly chart
Blockchain3 days ago

Coinbase Addresses Future Revenue Concerns With Plans to Become Crypto’s Amazon

Esports5 days ago

Zayt Retires From Competitive Fortnite For The Second Time

Esports4 days ago

Twitch streamer Lando Norris takes Italian F1 Grand Prix podium

Esports5 days ago

Dota 2: DPC Weekly Recap — SEA Apr 12-17, 2021

Esports4 days ago

Cloud9 Perkz says Kassadin can’t ever be balanced in LoL

Esports4 days ago

Three takeaways from the SWT Japan Ultimate Online Qualifier

Trending