Connect with us

Fintech

MIT Student Invents Breakthrough Art Restoration Technique | PYMNTS.com

Published

on

Ever since he was a child, Alex Kachkine has been fascinated by paintings. He would visit museums and was drawn in by the visual art depicted in landscapes, historical figures and religious scenes.

“Anytime I visit New York City, the first place I go to is the art gallery,” Kachkine said in an interview with PYMNTS. “It’s been a lifelong passion of mine.”

Such adoration naturally means Kachkine would look to acquire art works of his own. But with a limited budget, the MIT graduate researcher with a discerning eye instead bought damaged oil paintings he could restore.

“I ventured into art conservation around 10 years ago when I realized that you can’t buy a Monet reasonably,” Kachkine said. “But you can, even with the limited income I had back then, buy damaged paintings. And I realized that I could take one of those damaged paintings, restore it, and then I would have a really nice painting.”

Kachkine knew that restoration is manually laborious. The painting has to be cleaned of debris and any past restoration efforts have to be removed as well. Then, the damaged parts in paintings have to be manually painted while staying true to the artist’s style.

This typically means months to years of painstaking work. Kachkine did it the traditional way at first, but thought there must be a better way. So, he invented a method using artificial intelligence (AI), transfer paper, printers and varnish. His paper describing the technique is published in the journal Nature.

Kachkine said his method greatly speeds up restoration: In repairing a 2-foot by 2-foot painting, “The Adoration of the Shepherds,” from the late 15th century, he spent 3.5 hours compared to 232 hours it would normally take to do it manually. That’s faster by 66 times.

Source: “Physical restoration of a painting with a digitally constructed mask,” Nature

Taking the cleaning time into account, his method would speed up the entire restoration process by four to five times, Kachkine said.

Around 70% of paintings in institutional collections are not displayed in public due in part of the cost of restoring them, according to Kachkine’s paper. Therefore, restoration efforts typically center around the most valuable pieces of art with the rest left buried in storage.

Kachkine said various AI models are able to generate images of damaged paintings as they would look fully restored. But these would exist only virtually. He said his technique is the first to translate the digital restored image into physically restoring the actual painting.

“This is the first time we’ve been able to take all of those digital tools and actually end up with a physically restored painting from them,” he said. “And it’s so much faster than doing these kinds of restorations by hand.”

How Gen AI Helps Restore Paintings

The process begins with cleaning the artwork of debris and old restoration efforts. Once cleaned, the painting is scanned to produce a high-resolution image. Kachkine then uses a variety of Adobe-integrated digital tools, including convolutional neural networks and partial convolution models, to reconstruct missing regions.

Once the digital restoration is complete, a transparent film mask is printed with the reconstructed imagery. This laminate consists of nine ultra-thin layers, including a white backing for color vibrancy and laser-printed pigments. The result is an overlay that sits precisely on the original painting, with printed colors covering only the damaged areas.

“It’s thinner than human hair,” Kachkine said, adding that the film is removable using standard conservation solvents, preserving the artwork underneath.

The ethical implications of this method were also central to Kachkine’s design. He developed algorithms that determine which regions to restore based on how human vision perceives color and contrast.

“We really only select the damages that human vision is sensitive to,” he said. “You can tell what areas have been restored and which have not. That’s really important from an ethical standpoint in conservation.”

At first, Kachkine said he wasn’t sure how his method would be received. But he was gratified to see broad interest from conservators, cultural institutions and private equity firms. He also has a GoFundMe page.

Kachkine said he is now collaborating with the Italian Ministry of Culture on restoring frescoes in earthquake-damaged chapels in Tuscany.

His dream painting restoration job would come from the Italian Renaissance.

“There are a number of Italian paintings, especially around the Renaissance, that have very bright colors” such as Raphael, Kachkine said. “I’d love to be able to restore one of those [paintings] where before restoration, it would be very difficult to appreciate all of the fun colors that might emerge and the interesting textures that are there.”

“That’s the dream,” he said. “It might take a little bit before I could get my hands on one, but I’ll keep trying.”

Read more:

New AI Startup Turns a Dog’s Bark Into a Human Conversation

Meta Says Foundation Model Gives Virtual Embodied Agents Human-Like Movements

AI-Powered Digital Twins Give Clinical Trials a 75-Year Upgrade

Photo: MIT graduate researcher Alex Kachkine looking at a painting. Credit: Alex Kachkine

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Fintech

SEC Forms Task Force Promoting ‘Responsible AI Integration’ | PYMNTS.com

Published

on

The initiative, announced Monday (Aug. 4), is designed to promote responsible use of AI while enhancing innovation and efficiency in the SEC operations. Valerie Szczepanik, who has been named the SEC’s chief AI officer, will head the task force.

“Recognizing the transformative potential of AI, the SEC’s AI Task Force will accelerate AI integration to bolster the SEC’s mission,” the regulator said in a news release.

“It will centralize the agency’s efforts and enable internal cross-agency and cross-disciplinary collaboration to navigate the AI lifecycle, remove barriers to progress, focus on AI applications that maximize benefits, and maintain governance. The task force will support innovation from the SEC’s divisions and offices and facilitate responsible AI integration across the agency.”

Before being named the chief AI officer, Szczepanik directed the SEC’s Strategic Hub for Innovation and Financial Technology. She has also served as associate director in the SEC’s Division of Corporation Finance a Special Assistant United States Attorney at the United States Attorney’s Office for the Eastern District of New York, according to the release.

The announcement comes two weeks after the White House released a policy roadmap outlining President Trump’s push to keep America in the lead in the global AI race.

“America’s AI Action Plan” follows Trump’s executive order in January that ordered federal agencies to overturn AI regulations put in place by the Biden administration, which focused on oversight and risk mitigation.

“As our global competitors race to exploit these technologies, it is a national security imperative for the United States to achieve and maintain unquestioned and unchallenged global technological dominance,” Trump said in the opening of the AI action plan.

In other AI news, recent research by PYMNTS Intelligence finds that almost all chief product officers (CPOs) expect generative AI to reshape the way they work.

That research showed that nearly all product leaders say AI will streamline workflows within three years, compared to 70% last year. And more than 80% anticipate improvements in data security, compared to half of the CPOs surveyed last year.

“The shift over the past year among CPOs reflects a deeper change in institutional mindset. Gen AI is no longer experimental — it’s strategic,” PYMNTS wrote. “The pressure to deliver more with fewer resources has pushed firms to scale automation of routine, labor-intensive tasks, not just explore how that can be done.”

Continue Reading

Fintech

Experian Unveils New AI Tool for Managing Credit and Risk Models | PYMNTS.com

Published

on

Experian Assistant for Model Risk Management is designed to help financial institutions better manage the complex credit and risk models they use to decide who gets a loan or how much credit someone should receive. The tool validates models faster and improves their auditability and transparency, according to a Thursday (July 31) press release.

The tool helps speed up the review process by using automation to create documents, check for errors and monitor model performance, helping organizations reduce mistakes and avoid regulatory fines. It can cut internal approval times by up to 70% by streamlining model documentation, the release said.

It is the latest tool to be integrated into Experian’s Ascend platform, which unifies data, analytics and decision tools in one place. Ascend combines Experian’s data with clients’ data to deliver AI-powered insights across the credit lifecycle to do things like fraud detection.

Last month, Experian added Mastercard’s identity verification and fraud prevention technology to the Ascend platform to bolster identity verification services for more than 1,800 Experian customers using Ascend to help them prevent fraud and cybercrime.

The tool is also Experian’s latest AI initiative after it launched its AI assistant in October. The assistant provides a deeper understanding of credit and fraud data at an accelerated pace while optimizing analytical models. It can reduce months of work into days, and in some cases, hours.

Experian said in the Thursday press release that the model risk management tool may help reduce regulatory risks since it will help companies comply with regulations in the United States and the United Kingdom, a process that normally requires a lot of internal paperwork, testing and reviews.

As financial institutions embrace generative AI, the risk management of their credit and risk models must meet regulatory guidelines such as SR 11-7 in the U.S. and SS1/23 in the U.K., the release said. Both aim to ensure models are accurate, well-documented and used responsibly.

SR 11-7 is guidance from the Federal Reserve that outlines expectations for how banks should manage the risks of using models in decision making, including model development, validation and oversight.

Similarly, SS1/23 is the U.K. Prudential Regulation Authority’s supervisory statement that sets out expectations for how U.K. banks and insurers should govern and manage model risk, especially in light of increasing use of AI and machine learning.

Experian’s model risk management tool offers customizable, pre-defined templates, centralized model repositories and transparent internal workflow approvals to help financial institutions meet regulatory requirements, per the release.

“Manual documentation, siloed validations and limited performance model monitoring can increase risk and slow down model deployment,” Vijay Mehta, executive vice president of global solutions and analytics at Experian, said in the release. With this new tool, companies can “create, review and validate documentation quickly and at scale,” giving them a strategic advantage.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

Read more:

Experian and Plaid Partner on Cash Flow Data for Lenders

Experian Targets ‘Credit Invisible’ Borrowers With Cashflow Score

CFPB Sues Experian, Alleging Improper Investigations of Consumer Complaints

Continue Reading

Fintech

Anthropologie Elevates Maeve in Rare Retail Brand Launch | PYMNTS.com

Published

on

Anthropologie is spinning off its Maeve product line as a standalone brand, a rare move in a retail sector where brand extensions have become less common.

The decision reflects shifting strategies among specialty retailers as they work to adapt to changes in women’s fast-fashion and evolving consumer behavior.

Maeve, known for its blend of classic silhouettes and modern flourishes, will now operate independently with dedicated storefronts and separate digital channels, including new social media accounts and editorial content platforms, according to a Monday (Aug. 4) press release. The brand is inclusive, spanning plus, petite, tall and adaptive options, which broaden its reach as the industry contends with demands for representation.

Maeve has nearly 2 million customers and was the most-searched brand on the Anthropologie website over the past year, the release said. It is also a driver of TikTok engagement. Several of the company’s most “hearted” items online are already from the Maeve label.

“Maeve has emerged as a true driver of growth within Anthropologie’s portfolio,” Anu Narayanan, president of women’s and home at Anthropologie Group, said in the release. “Its consistent performance, combined with our customers’ emotional connection to the brand, made this the right moment to evolve Maeve into a standalone identity.”

While many retailers have retreated from new brand creation, opting instead to consolidate or focus on core labels, Anthropologie’s move suggests confidence in cultivating sizable, engaged consumer communities around sub-brands.

Anthropologie is backing Maeve’s standalone debut with a comprehensive marketing campaign, including influencer-driven content, a new Substack, a launch event in New York, and a charitable partnership, per the release. The first Maeve brick-and-mortar store is set to open in Raleigh, North Carolina, in the fall.

The move comes as the apparel sector in the United States sees shoppers valuing not just price and selection, but brand story, inclusivity and digital experience. While the outcome remains to be seen, Anthropologie’s gamble on Maeve reflects a belief that consumers remain eager to embrace distinctive, thoughtfully curated fashion.

Continue Reading

Trending