Connect with us

Technology

Dnotitia Unveils STAR-KV, Achieving UP to 20x KV Cache Compression, Selected as an ICML 2026 Spotlight Paper

Published

2 months ago

on

Introduces a low-rank-based approach to KV cache compression, one of the key bottlenecks in long-context AISpeeds up attention computation by up to 6.9x and overall generation throughput by up to 3.1x, moving beyond memory savings to faster inferenceSelected as a Spotlight paper at ICML 2026, representing about 2.2% of reviewed submissions and about 8.4% of accepted papersFollowing the attention around Google’s TurboQuant at ICLR 2026, STAR-KV presents another approach to advancing KV cache compressionPaper available on arXiv; source code released on GitHub

SEOUL, South Korea, July 1, 2026 /PRNewswire/ — Dnotitia Inc. (Dnotitia), a company specializing in long-term memory AI and semiconductor-based AI infrastructure technologies, has released the paper and source code for “STAR-KV: Low-Rank KV Cache Compression via Soft Thresholding for Adaptive Rank Control.” The technology was developed through a joint research effort involving UC San Diego’s VVIP Lab and Dnotitia researchers, and the paper was selected as a Spotlight paper at ICML 2026 (International Conference on Machine Learning 2026), one of the world’s leading conferences in machine learning.

In the experiments reported in the paper, low-rank compression alone reduced the KV cache by up to 75%. Combined with the mixed-precision quantization method proposed in the paper, STAR-KV compressed the full KV cache by up to 20x. The technology also improves computation speed through custom GPU kernels, increasing attention computation speed by up to 6.9x and overall generation throughput by up to 3.1x. STAR-KV also showed higher accuracy than major existing KV cache compression methods.

KV cache compression has become a key technical challenge in AI infrastructure. As research into reducing the memory bottleneck of long-context AI gains momentum, including the attention around Google’s TurboQuant at ICLR 2026, STAR-KV presents a new approach that combines low-rank compression with quantization and GPU execution optimization.

The KV cache is temporary memory stored on the GPU so that a large language model (LLM) does not have to recompute context it has already processed. As AI evolves into agentic systems that use multiple documents, conversation history, code, search results, and outputs from external tools, the amount of context a model must process is growing rapidly. In this environment, the KV cache has emerged as a key bottleneck affecting both GPU memory usage and inference cost.

According to the STAR-KV paper, when a LLaMA-3.1-8B model processes a 128K-token context at a batch size of 4, the KV cache accounts for about 81% of total GPU memory. As long-context AI becomes more widely used, KV cache compression is increasingly viewed as a core AI infrastructure technology for processing long context at lower cost.

ICML, where the STAR-KV paper was accepted, is widely regarded as one of the top international conferences in AI and machine learning, alongside NeurIPS and ICLR. ICML 2026 will be held from July 6 to 11 at COEX in Seoul. This year, 23,918 papers entered review, 6,352 were accepted, and 536 were selected as Spotlight papers. Spotlight papers account for about 2.2% of all reviewed submissions and about 8.4% of accepted papers.

Going forward, Dnotitia plans to further advance STAR-KV for use in real-world AI service environments and explore its application to open-source LLM inference frameworks such as vLLM.

“Technologies that help AI process longer context faster and at lower cost are advancing rapidly” said MK Chung, CEO of Dnotitia. “STAR-KV addresses the core bottlenecks in KV cache capacity and attention processing speed, and Dnotitia aims to contribute to the AI inference ecosystem through open sourcing.”

View original content to download multimedia:https://www.prnewswire.com/news-releases/dnotitia-unveils-star-kv-achieving-up-to-20x-kv-cache-compression-selected-as-an-icml-2026-spotlight-paper-302815857.html

SOURCE Dnotitia Inc.

Related Topics:
Continue Reading
Click to comment

Leave a Reply

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

Technology

Fastest Growing Global Finance Super App Abound chooses Monad to Power its Next Phase of Growth

Published

18 minutes ago

on

August 25, 2026

By

Riding record growth after crossing $500 Million in remittances, the super-app built for global Indians announces its next chapter – including a brand-new wallet product, with the waitlist opening soon

SAN FRANCISCO, Aug. 25, 2026 /PRNewswire/ — Abound, the super-app for Indians in the US backed by Times Internet (Times of India Group), grew past $500 Million in remittance volumes served in the last year, with US and India markets accounting for the bulk of volumes. Abound is excited to announce the next phase of products being built to serve the global Indian diaspora better. The next chapter will be built on Monad, the next-generation Layer 1 blockchain enabling open access to fair financial markets. With Monad’s technology, it is the best option for Abound’s use cases.

What began as a smarter way to send money home has evolved into a comprehensive super app for NRI life spanning cross-border payments, savings, and everyday benefits designed around the needs of Indians living in the US. Abound has become one of the platforms Indians in the US trust to move and manage money across two countries, with over $500 million in volume served. Abound now plans to serve Indians across the world and expand support with newer products better suited to their changing needs.

A New Roadmap for the global Indian

Building on this momentum, Abound today unveiled the next chapter of its product vision:

A new wallet product: a single home for Indians overseas to hold, move, and manage money across both countries’ multi-currency balances, and instant transfers between US and India accounts, with support for more countries coming soon. Waitlist opens soon; the product launches in Q3 2026.New ways to earn and save: interest-bearing savings and investment accounts, giving NRIs more ways to put their money to work.AI agent-led workflows, including the agentic payment protocol x402 to power the next chapter of Abound’s upcoming product suite. Monad Foundation is a premier member of the x402 Foundation.

“Abound has become the platform Indians in the US trust to move money home; we have moved $500 million since launch” said Nishkaam Mehta, CEO, Abound. “This roadmap is about building agents to seamlessly manage the NRIs entire financial life, with the trust of our community at the center of everything we do.”

“Abound is embarking on an ambitious journey to build a full set of agentic financial services and cross border platform for NRIs along its distribution via Times of India. We’re excited to see Abound leverage Monad’s technology to help accelerate at their efforts into the next chapter,” said Raj Parekh, Head of Payments & Stablecoins.

Powered by Next-Generation Technology

Abound’s upcoming product suite will be built on Monad. Monad is a high-performance, institutional-grade Layer 1 blockchain enabling open access to rewards, AI agents, and fair financial markets on neutral ground. The integration gives Abound the technology foundation to serve the Indian diaspora at scale – with the performance and reliability a community of global earners expects. Monad’s technology underpins Abound’s expanded capabilities as the platform scales into its next chapter.

 

View original content to download multimedia:https://www.prnewswire.com/news-releases/fastest-growing-global-finance-super-app-abound-chooses-monad-to-power-its-next-phase-of-growth-302859236.html

SOURCE Abound

Continue Reading

Technology

Xinhua Silk Road: Seven marine test “labs” put in place in offshore waters near Yuanyao port in Weihai, E China

Published

18 minutes ago

on

August 25, 2026

By

BEIJING, Aug. 25, 2026 /PRNewswire/ — In waters north of Yuanyao fishery port in east China’s Weihai City, a recent round of drone marine testings by a local marine equipment institute pooled firsthand data for references to its product R&D.

Such testings that move corporate “laboratories” into the sea are increasingly favored by businesses in Weihai as seven marine testing platforms have been completed in the port area, also the Yuanyao shallow sea sci-tech bay area.

In details, the seven platforms include an offshore self-elevating wharf, a semi-submersible platform, a testing wharf, etc., all of which provide vital support to the technological R&D of related businesses.

Currently, 44 companies have been attracted to settle in the sci-tech bay area, which serves as a key carrier for emerging marine-related industries such as marine electronic information, marine intelligent equipment, and modern maritime services to thrive there.

Near the Yuanyao fishery port now stands the semi-submersible marine test platform, which is the first one of the type in Shandong Province where Weihai sits and also a core part of the national marine comprehensive test site project in Weihai.

As a staff of the sci-tech bay area’s service center introduced, the 44-meter-long and 32.4-meter-wide platform can operate at a maximum water depth of 70 meters and is equipped with multiple supporting systems.

For instance, the power system, test facilities, marine environment observing and monitoring system, and security system of the platform are capable of catering to diverse needs of scientific researchers including short-term on-site operations, instrument lifting and retrieval, and real-time transmission of offshore data.

Apart from these marine testing platforms, the sci-tech bay area is also crafting another seven scientific research platforms to offer all-around intellectual support for the national marine comprehensive test site in Weihai.

While marine testings are ongoing on the testing platforms, onshore zones in the sci-tech bay area are busy with building more innovative infrastructure including the marine data management and application center.

In Weihai, the integrated deployment of offshore, onshore and on-island marine-related platforms is unlocking strong potential for local industries to burgeon.

Original link: https://en.imsilkroad.com/p/351893.html

 

View original content:https://www.prnewswire.com/apac/news-releases/xinhua-silk-road-seven-marine-test-labs-put-in-place-in-offshore-waters-near-yuanyao-port-in-weihai-e-china-302859240.html

SOURCE Xinhua Silk Road

Continue Reading

Technology

Antare launches AI physical security platform, transforming unwatched camera footage into intelligence

Published

18 minutes ago

on

August 25, 2026

By

Antare uses AI to automatically review footage from body-worn and fixed cameras to give organisations a complete picture across every shift and site

LONDON, Aug. 25, 2026 /PRNewswire/ — Antare, a new agentic security platform for body-worn and fixed cameras, has launched. UK-based Antare uses artificial intelligence to automatically review security footage, detect incidents, alert teams and document every event, transforming everyday operations into actionable insight that helps improve performance and reduce operational risk. The platform is compatible with both Antare’s intelligent body-worn cameras and customers’ own existing fixed camera infrastructure.

Antare’s body-worn cameras capture the moments before, during and after every incident, preserving the context that helps organisations understand what really happened, protect their employees, and manage their liabilities. This matters because abuse and violence towards frontline staff is rising and incidents often do not have a defined beginning. Employees no longer have to remember to start a recording in the middle of a confrontation.

The Antare Console reviews all video and audio as it is recorded, and transcribes, indexes and scores each event automatically. Separate AI agents watch for specific signals, including physical and verbal assaults, suspicious behaviour, safety risks, medical events or weapons, and send in-the-moment alerts and live video to colleagues via the Antare App.

Antare also surfaces patterns that manual review is prone to miss, like repeat offenders, theft hotspots and recurring problems across sites. Free text search allows users to investigate further, while daily email summaries ensure key events are never missed.

Available today

Antare offers two models of body-worn cameras that work out of the box for US, UK, and EU customers. The Antare Body-worn Camera Pro is designed for frontline staff in demanding environments where extended battery life, GPS capabilities and rugged durability are important.

The Antare Body-worn Camera Compact is a lightweight alternative, intended for customer-facing teams who need discreet protection during their shifts. The Antare Dock charges four cameras at once between shifts.

Both models, including cellular connectivity and unlimited users for the Antare Console, are available through a simple, all-inclusive monthly subscription, with no minimum order quantity and no fixed term.

One platform, every camera

The Antare Cloud Gateway, coming soon, integrates the full range of Antare’s cloud capabilities with customers’ existing fixed cameras, so body-worn and fixed cameras live in a single console, with no separate systems to manage.

Customers can also choose from a range of pre-configured alarms, as well as configure their own alarms via free text description.

Antare’s own fixed camera range is set to follow in early 2027, supplied and installed by accredited partners. Partners can register interest now at www.antare.ai/partners.

Privacy and data protection

Footage is encrypted on Antare devices and in transit, then erased from the device once it reaches the secure cloud, where access is role-based. Data is stored in your local region – UK, EU or US – and deleted automatically after 30 days unless an authorized user marks it for retention. Antare is SOC 2 Type I certified and GDPR compliant, with SOC 2 Type II and ISO 27001 certifications in progress.

Mark Michaelides, CEO of Antare:

“Antare was founded on the belief that today’s physical security platforms are failing to deal effectively with emerging safety and operational risks, leaving businesses with expensive legacy investments that do not fulfil their evolving needs.

“Traditional security solutions have focused on after-the-fact review, while more modern systems have enabled live alerts. Antare has built a different type of intelligence platform to deliver both, while shortening time from alert to triage and enabling quicker post-incident investigation.

“Following strong results in trials across private security, hospitality and retail customers, Antare is available on subscription for body-worn cameras from today, with the Antare Cloud Gateway coming soon. One console that turns disparate data into action, intelligence, and insights.”

For more information, visit www.antare.ai.

Accompanying photography:

Picture 1 – Antare CEO Mark Michaelides. Credit: Antare 
Picture 2 – Credit: Antare 
Picture 3 – Credit: Antare 
Picture 4 – Credit: Antare 
Picture 5 – Credit: Antare 
Picture 6 – Credit: Antare 
Picture 7 – Credit: Antare

About Antare

Antare is the physical security intelligence platform that turns every captured event into organisational insight. It continuously detects, reviews and interprets everything that happens across every site and operation – so nothing is missed, and every incident is clearly understood.

Starting with body-worn and fixed cameras, and expanding across every capture point, Antare gives organisations the confidence to report accurately, respond effectively, and build a complete picture of risk and performance over time. Antare is headquartered in the United Kingdom. For more information, visit www.antare.ai.

View original content to download multimedia:https://www.prnewswire.com/news-releases/antare-launches-ai-physical-security-platform-transforming-unwatched-camera-footage-into-intelligence-302859242.html

SOURCE Antare

Continue Reading

Trending