Connect with us

Press Release

Major Advance in Lightweight and Privacy-Preserving NLP: EmByte Achieves High Accuracy Using Only 1/10Embedding Memory

Published

3 months ago

on

Brunswick, New Jersey, 23rd January 2026, ZEX PR WIRE, A newly published study in the Findings of the Association for Computational Linguistics: EMNLP 2025 introduces EmByte, a natural language processing (NLP) model that dramatically reduces embedding memory usage while improving accuracy and strengthening privacy protections. Developed by Jia Xu Stevens and collaborators, EmByte demonstrates that modern language models can operate with approximately 1/10 of the embedding memory used by conventional subword-based systems, while also achieving better task accuracy and up to 3-fold improvements in privacy resistance.

The EMNLP 2025 Findings paper presents EmByte as a byte-level embedding framework that replaces large subword vocabularies with compact, decomposed representations. This design significantly reduces the memory footprint of embedding layers—traditionally one of the largest components of NLP models—without increasing sequence length or computational overhead.


Small Embeddings, Strong Results

Embedding tables in standard NLP models often contain tens or hundreds of thousands of entries, consuming large amounts of memory and posing privacy risks when exposed to inversion or reconstruction attacks. EmByte addresses these challenges by representing text at the byte level and applying a decomposition-and-compression learning strategy that preserves semantic information while occupying much less space.

Experimental results reported in the EMNLP 2025 Findings paper show that EmByte:

  • Uses about 5% of the embedding memory required by typical subword models

  • Matches or exceeds accuracy on benchmark tasks such as classification, language modeling, and machine translation

  • Provides significantly stronger privacy protection, making it substantially harder to reconstruct original text from embeddings or gradients

These results demonstrate that embedding size reduction does not require sacrificing model quality. Instead, careful design of the representation can improve both performance and security.

Privacy by Design

A key contribution of EmByte is its impact on privacy. Because byte-level embeddings avoid direct one-to-one mappings between tokens and semantic units, they reduce the amount of recoverable information stored in each vector. This makes common attacks—such as embedding inversion and gradient leakage—far less effective.

According to the EMNLP 2025 Findings results, EmByte’s structure provides roughly three times stronger resistance to privacy attacks than standard embedding approaches. This makes the model especially relevant for sensitive domains such as healthcare, finance, and personal communications, where data protection is critical.

Built on a Long Line of Research

The EmByte framework builds directly on Jia Xu Stevens’s long trajectory of researchin efficient text representation, segmentation, and multilingual processing. Earlier work laid the conceptual and technical foundations for compact and robust language modeling, including:

  • Research on byte-based and subword modeling for multilingual and low-resource settings (EMNLP 2020; COLING 2022)

  • Studies on Chinese word segmentation and synchronous modeling that emphasized efficient representation and structural alignment

  • Early work in machine translation and speech-to-text processing that explored minimal and adaptive linguistic units

Together, these contributions reflect a consistent research direction: reducing redundancy in language representations while improving robustness, generalization, and security.

Implications for Real-World AI

By drastically reducing the memory requirements for embedding, EmByte enables the deployment of capable NLP models in environments with strict memory and privacy constraints. This includes:

  • On-device and edge AI systems

  • Privacy-sensitive enterprise and government applications

  • Large-scale systems where embedding tables dominate memory cost

EmByte also aligns with a broader shift in AI research away from purely scaling model size and toward architectural efficiency and responsible design.

Looking Forward

With its publication in Findings of EMNLP 2025, EmByte is positioned to influence future work on embedding design, privacy-preserving NLP, and efficient language models. The results suggest that smaller, more secure representations can outperform larger ones when designed with structure and learning dynamics in mind.

As language models continue to be integrated into everyday technology, approaches like EmByte point toward a future in which accuracy, efficiency, and privacy improve together rather than compete.

About Jia Xu Stevens

Jia Xu Stevens is a researcher in natural language processing and machine learning whose work spans efficient language representation, multilingual modeling, privacy-preserving AI, and text segmentation. Over the course of her research career, Jia Xu Stevens has contributed foundational and applied work across multiple generations of NLP systems, from early machine translation and word segmentation frameworks to modern embedding compression and privacy-aware language models.

Her research has been published at leading international venues, including EMNLP, COLING, IWSLT, and other ACL-affiliated conferences. A recurring theme in her work is the design of compact, structured language representations that improve robustness, generalization, and efficiency while reducing memory usage and privacy risks. This line of research includes early studies on synchronous segmentation and translation, later advances in subword and byte-based modeling, and recent innovations in embedding compression and privacy resistance.

Jia Xu Stevens’ work emphasizes architectural efficiency over brute-force scaling, demonstrating that carefully designed representations can outperform larger models while enabling safer real-world deployment. Her recent research continues to focus on building language technologies that are accurate, lightweight, and privacy-conscious, with applications ranging from multilingual NLP to on-device and resource-constrained AI systems.

About Author

Cloud PR Wire

See author's posts

Disclaimer: The views, suggestions, and opinions expressed here are the sole responsibility of the experts. No Digi Observer journalist was involved in the writing and production of this article.

Related Topics:
Continue Reading

Press Release

Trakx expands Canton presence as CTIs become available through Five North’s Loop wallet

Published

16 hours ago

on

April 18, 2026

By

Trakx, the leading platform for crypto-index trading, announced that its Crypto Tradable Indices (CTIs) are now available through Five North’s Loop wallet on Canton, enabling CTIs to be held and transferred on-chain through a Canton-native wallet environment. The milestone marks a new step in the evolution of Trakx’s index products beyond the Trakx platform and into the broader infrastructure of the Canton ecosystem.

With this integration, Trakx Crypto Tradable Indices gain a new layer of accessibility and portability on-chain. By becoming available through Loop, Five North’s wallet product for Canton, CTIs can now be held and transferred across the Canton Network, an institutional-grade blockchain environment designed for privacy and interoperability.

The development is strategically significant for Trakx as it continues to strengthen the infrastructure and distribution layer surrounding its crypto index products. While CTIs have historically been accessed primarily through the Trakx platform, their availability through Loop reflects a broader shift toward interoperability, partner integration, and professional distribution models across digital assets.

“This milestone is about more than wallet availability,” said Lionel Rebibo, CEO at Trakx. “It reflects the direction we believe digital asset products need to take: beyond standalone platform access and toward deeper integration within institutional-grade blockchain infrastructure. By making CTIs available through Loop on Canton, we are taking a concrete step toward broader B2B and B2B2C distribution, stronger interoperability, and a more scalable framework for partners looking to integrate structured digital asset exposure into their own environments.”

Canton Network plays an important role in that strategy. Designed as a privacy-enabled blockchain network with a strong institutional orientation, Canton provides the type of infrastructure environment that can support on-chain issuance, asset portability, and interoperable financial workflows. For Trakx, this creates a foundation for CTIs to evolve from platform-native products into assets that can be integrated into broader access and distribution models.

Five North‘s role in the Canton ecosystem gives additional strategic relevance to the launch. As a builder and operator of core infrastructure on Canton Network, including wallets, explorers, and validator systems, Five North occupies a meaningful position within the network. Its Loop wallet provides the practical layer through which CTIs can now be held and transferred on-chain, helping translate tokenization into real ecosystem usability.

For Trakx, the move is especially relevant in the context of partner-led adoption. The company sees increasing value in enabling B2B platforms, B2B2C distributors and other professional intermediaries to access more structured and infrastructure-ready models for digital asset exposure. In that sense, the integration is not simply a technical enhancement, but part of a broader distribution strategy focused on making CTIs easier to integrate, distribute, and use across professional environments.

As digital asset markets mature, infrastructure, accessibility, and interoperability are becoming just as important as product design. With CTIs now available through Five North’s Loop wallet on Canton, Trakx is advancing its objective of building not only better crypto index products, but also better infrastructure pathways for those products to be held, transferred and integrated across the next generation of on-chain finance.

About Trakx

Trakx is a crypto index investing platform offering Crypto Tradable Indices designed to provide structured exposure to digital asset markets. Its mission is to make crypto investing simpler, more transparent, and more accessible through index-based products built for both individual and professional use.

Contact Information

Gary Rebibo | CMO at Trakx | gary@trakx.io

Website | LinkedIn | X/Twitter | Discord | Telegram

Media Contact

Organization: Trakx

Contact Person: Gary Rebibo

Website: https://trakx.io/

Email: Send Email

City: Paris

State: France

Country:France

Release id:44145

The post Trakx expands Canton presence as CTIs become available through Five North’s Loop wallet appeared first on King Newswire. This content is provided by a third-party source.. King Newswire makes no warranties or representations in connection with it. King Newswire is a press release distribution agency and does not endorse or verify the claims made in this release. If you have any complaints or copyright concerns related to this article, please contact the company listed in the ‘Media Contact’ section

file

About Author

Cloud PR Wire

See author's posts

Disclaimer: The views, suggestions, and opinions expressed here are the sole responsibility of the experts. No Digi Observer journalist was involved in the writing and production of this article.

Continue Reading

Press Release

General Compute Launches ASIC-First Inference Cloud for Autonomous AI Agents

Published

16 hours ago

on

April 18, 2026

By

General Compute today announced its inference cloud platform built for AI agents, working with early partners now ahead of general availability on May 15, 2026. The platform runs on purpose-built AI accelerators rather than general-purpose GPUs. More information is available at generalcompute.com.

SAN FRANCISCO — April 18, 2026 — General Compute Inc. today announced its inference cloud platform, which is designed for AI agent workloads. The company is working with early partners now, with general availability scheduled for May 15, 2026.

The platform runs on purpose-built AI accelerators rather than general-purpose graphics processors. Its architecture separates the prefill and decode stages of inference processing, allowing each stage to be scaled independently based on workload.

The platform is built to serve AI agents that make high volumes of LLM inference and tool calls, including AI agents that provision their own compute programmatically.

“The last 20 years we built for developers, the next 20 we will build for agents. On General Compute, AI agents can sign up on their own and provision their own inference. Our docs and API are optimized for both human and AI agent consumption,” said Jason Goodison, co-founder and Chief Technology Officer of General Compute.

Platform Overview

The platform offers an industry-standard API, allowing developers to integrate it into existing applications with minimal code changes. AI agents and developers alike can sign up, provision API keys, and begin making inference calls programmatically.

At launch, the platform will offer access to a range of open-source LLMs across multiple model families and parameter sizes. Customers can also deploy their own models on the company’s infrastructure.

Infrastructure

General Compute’s data center infrastructure operates on hydroelectric power. The company states that its accelerator hardware is air-cooled, and that its racks operate at lower power densities than comparable installations built on general-purpose processors.

The company publishes technical performance data for its platform on its website.

Availability

General Compute is working with early partners now, with general availability beginning May 15, 2026. Enterprise inquiries regarding dedicated infrastructure, service level agreements, and capacity planning may be directed to jason@generalcompute.com

About General Compute

General Compute Inc. is an inference cloud company headquartered in California. The company was founded by Jason Goodison and Finn Puklowski.

Contact

Jason Goodison, Co-founder and Chief Technology Officer General Compute Inc. jason@generalcompute.com generalcompute.com

Media Contact

Organization: General Compute Inc

Contact Person: Jason Goodison

Website: https://generalcompute.com

Email:
jason@generalcompute.com

Contact Number: +14257537667

Address:440 North Barranca Avenue

Address 2: 3780

City: Covina

State: California

Country:United States

Release id:44077

The post General Compute Launches ASIC-First Inference Cloud for Autonomous AI Agents appeared first on King Newswire. This content is provided by a third-party source.. King Newswire makes no warranties or representations in connection with it. King Newswire is a press release distribution agency and does not endorse or verify the claims made in this release. If you have any complaints or copyright concerns related to this article, please contact the company listed in the ‘Media Contact’ section

file

About Author

Cloud PR Wire

See author's posts

Disclaimer: The views, suggestions, and opinions expressed here are the sole responsibility of the experts. No Digi Observer journalist was involved in the writing and production of this article.

Continue Reading

Press Release

Tempo Mails Launches Temporary Email Address Platform for Instant Disposable Inbox Access

Published

1 day ago

on

April 18, 2026

By

MIAMI, Florida, United States — April 18, 2026 — Tempo Mails today announced the launch of its free temporary email address platform, designed to help users create disposable email addresses instantly for sign-ups, verification codes, online trials, and privacy-focused browsing. The web-based service provides a temporary mailbox without registration and supports real-time inbox access for users who need a burner email without linking it to a personal account.

Available at tempomails.com, the platform is built for users who want a faster way to manage one-time online interactions while keeping their primary inbox separate from promotional messages, verification traffic, and unwanted follow-up emails. According to the company’s website, the service creates a temporary email address on page load, supports inbox history controls, and allows users to refresh, delete, or add new addresses from the interface. 

Tempo Mails said the service is intended for common online use cases such as account verification, e-commerce sign-ups, app testing, temporary registrations, and limited anonymous communication. The platform also supports temporary email send and receive workflows where users need short-term inbox access to receive OTPs, confirmation links, and service-related messages without using a permanent account. The site states that no sign-up is required and that inboxes are auto-cleared after use. 

The service includes features such as custom email alias options, real-time inbox refresh, auto-delete functionality, and multi-domain address generation. Tempo Mails also states that the platform is free to use and designed to work worldwide, giving users access to a disposable inbox through a browser-based interface rather than a downloadable application. 

“With Tempo Mails, the goal is to make privacy-focused email access simple for everyday users,” said James at Tempo Mails. “People often need a temporary mailbox for a single task, such as receiving a code or completing a registration, without exposing their personal inbox to long-term marketing traffic.”

The launch comes as more users look for practical ways to separate account creation and one-time online activity from their primary email accounts. By offering disposable email addresses that can be created in seconds, Tempo Mails is positioning the platform as a lightweight privacy tool for consumers, testers, developers, and online shoppers who need a burner email for short-term use.

About Tempo Mails
Tempo Mails is a web-based temporary email service that provides free temporary email address generation for users who want short-term inbox access without registration. The platform offers disposable email addresses, real-time inbox refresh, custom alias controls, and temporary mailbox features designed for account verification, online trials, and privacy-focused browsing. The service is available through tempomails.com and is designed for fast access across devices.

Media Contact

Organization: Tempo Mails

Contact Person: James

Website: https://tempomails.com/

Email:
contact@tempomails.com

Country:United States

Release id:44144

The post Tempo Mails Launches Temporary Email Address Platform for Instant Disposable Inbox Access appeared first on King Newswire. This content is provided by a third-party source.. King Newswire makes no warranties or representations in connection with it. King Newswire is a press release distribution agency and does not endorse or verify the claims made in this release. If you have any complaints or copyright concerns related to this article, please contact the company listed in the ‘Media Contact’ section

file

About Author

Cloud PR Wire

See author's posts

Disclaimer: The views, suggestions, and opinions expressed here are the sole responsibility of the experts. No Digi Observer journalist was involved in the writing and production of this article.

Continue Reading

LATEST POST

Like Us
Follow Us
Subscribe Us
Follow Us