BEHAVOX
18.5.2022 15:51:04 CEST | Business Wire | Pressemeddelelse
Behavox , som udbyder en række sikkerhedsprodukter, der hjælper compliance-, HR- og sikkerhedsafdelingerne med at beskytte deres virksomhed og kolleger mod ondsindede aktører, fortæller i dag, at dens akademiske forskningsdokument om kunstig intelligens "kontinuerlig læring med lattice-fri MMI til stemmegenkendelse" er blevet accepteret af International Conference on Acoustics, Speech, and Signal Processing (ICASSP). ICASSP er verdens største og mest omfattende tekniske konference med fokus på signalbehandling og tilhørende anvendelse. Behavox præsenterede den 9. maj 2022 dette dokument i Singapore for IEEE ICASSP 2022 .
"Nogle gange forsker videnskabsfolk for videnskabens skyld, og samfundet oplever ikke nogen umiddelbare fordele af forskningen. Vi har derimod skubbet videnskaben til en ny grænse for at finde en løsning på et aktuelt problem," siger Behavox's CTO Joseph Benjamin. "Vi er beærede over at have modtaget anerkendelse fra en så prestigefyldt konference og er meget stolte af vores team og det arbejde, folkene har udført."
Dokumentet, der er forfattet af Behavox's maskinlæringsteam, ser på udfordringerne i forbindelse med løbende forbedring af stemmegenkendelsessystemer i miljøer, der kræver beskyttelse af personoplysninger. Vores forskning har direkte forbindelse til Behavox's FoU-tilgang, der giver kunderne den bedste transskriptionskvalitet på det finansielle marked. I FinTech sker der hele tiden ændringer i det sprog og de data, som Behavox-systemerne hurtigt og pålideligt skal behandle, og stigningen i kryptovalutaer og videokommunikation er blot to eksempler.
Derfor skal maskinlæringsmodeller jævnligt opdateres med relevante data. For at gøre sådanne opdateringer mulige, uden at sætte databeskyttelsen over styr, skal modellen gradvist kunne forbedres i sikre klientmiljøer uden adgang til historiske data og data fra andre kunder. Dette er en alvorlig udfordring inden for maskinlæring på grund af den katastrofale glemsomhed, hvor modellen, efter at have forbedret en bestemt type data, begynder at præstere dårligt på tidligere lærte data.
ICASSP-dokumentet fra Behavox bidrager til den fortsatte læring til brug ved stemmegenkendelse. Forfatterne foreslår en ny algoritme, der udvider og forfiner den almindeligt anvendte LWF-teknik (Learning Without Forgetting), der er afhængig af en regulariseringsterm og tvinger neurale netværk til at mindske glemsomhed. Mere specifikt designede forfatterne en nyhed i form af neurale netværk med indlæring af tab på sekvensniveau, der kan bruges i stedet for den konventionelle, punktvise LWF. De viste en betydelig reduktion i glemsomhed, når man finjusterede det neurale netværk sekventielt efter forskellige accenter- og taleformer.
Behavox har konstateret, at brugen af stemmeplatforme med årene er steget 94 procent, og at næsten en tredjedel af al kommunikation på arbejdspladsen i gennemsnit foregår som telefonmøder eller videokonferencer.
Denne teknik er allerede blevet anvendt med succes på dets danske system, hvor samarbejdet mellem Behavox's lingvister, maskinlæringsteamet og en kunde gav Behavox mulighed for at opbygge sit første danske transskriptionssystem, der arbejder pålideligt med bankdata.
Om Behavox
Behovox udbyder en række sikkerhedsprodukter, der hjælper compliance-, HR- og sikkerhedsafdelingerne med at beskytte deres virksomhed og kolleger mod ondsindede aktører.
Med AI-baseret analyse af alle virksomhedens kommunikationsdata, herunder e-mail, chat, telefonmøde- og videokonferenceplatforme, hjælper Behavox virksomhederne med at identificere ulovlig, umoralsk og skadelig adfærd på arbejdspladsen.
Behavox, der blev grundlagt i 2014, har hovedkvarter i Montreal og kontorer i New York, London, Seattle, Singapore og Tokyo.
Originalsprogsudgaven af denne bekendtgørelse er den officielle, autoriserede version. Oversættelserne er kun tænkt som en hjælp og bør sammenholdes med kildesprogsteksten, der som den eneste er juridisk bindende.
Se kildeudgaven på businesswire.com: https://www.businesswire.com/news/home/20220513005572/da/
Link:
Information om Business Wire
Følg pressemeddelelser fra Business Wire
Skriv dig op her, og modtag pressemeddelelser på e-mail. Indtast din e-mail, klik på abonner, og følg instruktionerne i den udsendte e-mail.
Flere pressemeddelelser fra Business Wire
Footprint Expands Into PE-Free Cups, Bringing a Proven, Recyclable and Home-Compostable Alternative to Plastic-Lined Paper Cups Into Production21.7.2026 18:00:00 CEST | Press release
Move strengthens Footprint’s European manufacturing platform and supports growing demand for practical alternatives to single-use plastic Footprint, a materials science and technology company focused on eliminating single-use plastics, today announced it has acquired a recyclable, home-compostable, polyethylene (PE)-free hot and cold cup platform from Transcend Packaging, along with next-generation paper lid and straw solutions. Transcend Packaging is a European company focused on innovation in sustainable packaging solutions. Together, the cup, lid, and straw expand Footprint’s portfolio with a proven solution for the global cup market. Footprint will begin scaling its new PE-free solution through its Poland facility as the company expands across Europe, with North America to follow. As operations ramp in Poland, Footprint will work with Transcend through a manufacturing partnership to support customer continuity across Europe and meet growing demand for practical alternatives to plas
Spectro Cloud Launches PaletteAI Inference Launchpad to Reduce Enterprise Token Costs by up to 70%21.7.2026 16:00:00 CEST | Press release
New turnkey, locally managed PaletteAI Inference Launchpad helps enterprises control token costs with local-first inference, while expanded PaletteAI support for AMD gives customers more choice across GPUs, models and AI infrastructure stacks. Spectro Cloud, a leading provider of AI infrastructure management software, today launched PaletteAI Inference Launchpad, a turnkey, locally managed solution designed to help enterprises reduce token costs by as much as 70% and run AI inference closer to their data, applications and users. This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20260721597271/en/ The company also announced expanded PaletteAI support for AMD-powered AI infrastructure, including AMD GPUs, AMD GPU Operator, ROCm™ runtime, and the AMD enterprise AI reference stack — including AMD-optimized models from the AMD Inference Microservices (AIMs) catalog. Together, the announcements expand PaletteAI as one platform for prod
RecVue Launches First AI Native Agentic Revenue Operating System21.7.2026 15:03:00 CEST | Press release
New agentic architecture for RevOS runs enterprise revenue across the order-to-cash lifecycle and surfaces cross-system intelligence RecVue, the leader in billing and revenue management solutions, today announced the evolution of RecVue RevOS into the industry's first Agentic Revenue Operating System. In an architectural shift, the now-native agentic revenue management platform moves revenue operations beyond AI-assisted workflows to fully governed, autonomous execution with more than 50 purpose-built agents that work across contracts, invoices and payment data to predict risk, surface anomalies and escalate exceptions throughout the quote-to-cash lifecycle. Enterprise revenue operations have long suffered from a structural gap between billing systems that generate invoices and ERP systems that record transactions. Finance teams face tedious, manual reconciliation month after month, often resulting in delayed cash, billing errors that generate disputes, revenue leakage from ungoverned
Yubico Extends Passkeys Beyond Trusted Authentication to Verified Authorization with Launch of YubiKey 5.821.7.2026 15:00:00 CEST | Press release
YubiKey 5.8 extends hardware-backed assurance from authentication into emerging digital signature, wallet, payment and agentic approval workflows Yubico (NASDAQ STOCKHOLM: YUBICO), the pioneer of phishing-resistant authentication and creator of the original passkey, the YubiKey, today announced the general availability of its YubiKey 5.8 – marking an expansion of the role of the passkey from secure authentication to include verifiable, hardware-backed authorization. The new firmware delivers a foundation for secure enterprise workflows across identity wallets, document signing, and AI-driven approvals, while also providing developers with the critical capabilities needed to test, design and deploy these next-generation security features early. This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20260721286409/en/ As generative and agentic AI mature to drive rapid, automated cyberattacks, traditional multi-factor authentication (MFA
Gurobi Publishes Inaugural State of Mathematical Optimization in Academia Report21.7.2026 15:00:00 CEST | Press release
Findings showcase how academics are leveraging optimization alongside GenAI tools in the classroom and in research today. Gurobi Optimization, LLC, the leader in decision intelligence technology, today announced the publication of its inaugural State of Mathematical Optimization in Academia Report. Based on surveys of over 1,180 faculty and students who actively use Gurobi, the report examines how optimization is being taught, used in research, and applied across academia today—as well as use cases that could benefit from a combination of generative AI (GenAI) with optimization. Gurobi’s research highlights the importance of making optimization education and tools accessible to students and instructors. “Mathematical optimization is powering some of the most effective decision-making across today’s enterprises, and the academic community is key to advancing its theory and practice,” said Duke Perrucci, CEO, Gurobi. “By exploring how students, researchers, and educators are teaching and
I vores nyhedsrum kan du læse alle vores pressemeddelelser, tilgå materiale i form af billeder og dokumenter samt finde vores kontaktoplysninger.
Besøg vores nyhedsrum
