Superposition refers to encoding representations of multiple features within a single neuron, which is common in deep neural networks. This property allows neurons to combine and represent multiple features, enabling the model to capture intricate information and handle complex tasks. Despite promising performance, the model's interpretability has been diminished. This paper presents a novel approach to enhance model interpretability by regularizing feature superposition. We introduce SAFR, which simply applies regularizations to the loss function to promote monosemantic representations for important tokens while encouraging polysemanticity for correlated token pairs, where important tokens and correlated token pairs are identified via VMASK and attention weights respectively. We evaluate SAFR with a transformer model on two classification tasks. Experiments demonstrate the effectiveness of SAFR in improving model interpretability without compromising prediction performance. Besides, SAFR provides explanations by visualizing the neuron allocation within the intermediate layers.

"We decided on SAFR, Realnetworks' AI facial recognition, because of its near 100 per cent accurate recognition," said Anthony Brooks, managing director at Geutebruck Pacific. "This accuracy allows Geutebruck video recordings to be analysed even more accurately, faster and automatically." Security operators will be supported in immediately recognising unwanted persons in restricted spaces by triggering an alarm within milliseconds. The access of unauthorised persons can also be identified reliably in recordings. "With SAFR, we are pleased to offer our customers this exceptional AI component that efficiently enhances an organisations safety and security," Brooks said.

Safr from RealNetworks has earned its third Small Business Research Innovation (SBIR) deal from the United States Air Force (USAF) to enable the extension of Safr AI-powered analytics, including facial recognition, to unmanned ground vehicles (UGVs). The UGVs would be used to reduce risks in perimeter protection and domestic emergency medical services (EMS) search and rescue missions. In a news release, the company said the contract will help improve its platform to be able to operate on an NVIDIA Jetson AGX Xavier-based UGV system, with the goal of reducing the risk service members face as the Safr-enhanced UGVs will be able to detect unauthorized persons in restricted areas with face biometrics. "As a USAF military working dog handler, I have employed canines in various environments fulfilling the multi-use role of detection and deterrence. The ability to utilize UGV systems to augment K9 teams during work/rest cycles, or as an additional force, broadens security in-depth and allows operations to continue unhindered," said Air Force Technical Sergeant Dustin Cain, Non-Commissioned Officer in Charge of Police Services, 366th Security Forces Squadron, Mountain Home Air Force Base, Idaho.

The premarket hours are back in the red again on Wednesday. But we're seeing plenty of penny stocks shine. The fact that most small-cap stocks are disconnected from the overall market trend is something I think is overlooked. While not all penny stocks will respond this way, many do. We saw this today with several of the breakout, midstream oil stocks that were moving hugely in after-hours trading on Tuesday evening.

SAFR from RealNetworks, Inc., the world's premier facial recognition and computer vision platform for live video, announced the addition of Walter Candelu as Area Vice President for the Middle East. Mr. Candelu brings experience and leadership to the new SAFR office in Dubai. He will drive its growing sales and business development initiatives across the Middle East Region. Mr. Candelu will be based in Dubai, UAE and will focus on expanding the SAFR reseller channel, partner network, and regional sales and marketing programs. Prior to joining SAFR, he held senior positions with leading security companies where he successfully drove the exponential growth of their technical capabilities and revenue.

Digital media software company RealNetworks has announced significant financial progress in the third quarter financial report, the company says in a press release. RealNetworks had already announced great success for its SAFR biometric facial recognition software in its Q2 2019 report that saw revenues of $44.2 million. According to current figures, the third quarter revenue rose to $45.0 million from just $17.6 million during the same timeframe last year. The SAFR live video artificial intelligence and biometric facial recognition platform was recently integrated with the Nvidia Metropolis Software Partner Program to improve technology deployment in smart city projects. In June, SAFR biometric facial recognition was rolled out in a museum in Europe on MOBOTIX cameras to provide live analytics.

WIRED recently highlighted unacceptable levels of bias in facial recognition in the article The Best Algorithms Struggle to Recognize Black Faces Equally. They cited the poor test scores of leading facial recognition vendors, as reported by the National Institute of Standards and Technology (NIST) in its July 2019 results. WIRED specifically called out Idemia but generalized their concerns. "The NIST test challenged algorithms to verify that two photos showed the same face, similar to how a border agent would check passports. At sensitivity settings where Idemia's algorithms falsely matched different white women's faces at a rate of one in 10,000, it falsely matched black women's faces about once in 1,000 -- 10 times more frequently. A one in 10,000 false match rate is often used to evaluate facial recognition systems."

With the wave of school shootings that have swept the U.S. in recent years, concerns about physical security and safety have overwhelmed parents, teachers and school administrators alike. Facial recognition technology, which would allow schools and law enforcement to quickly identify who is entering their schools and when could give school districts a powerful means to make schools even safer. Last month, RealNetworks, the streaming media company that garnered attention in the '90s and early 2000s for developing the first audio streaming solution, announced it would offer its facial recognition software, SAFR, for free to over 100,000 school districts. "School safety has become one of the top national issues in the United States in 2018," said Rob Glaser, chairman and CEO of RealNetworks in a press release. "We are proud to give our leading-edge SAFR for K-12 technology solution to every elementary, middle, and high school in America and Canada. We hope this will help make schools safer."

SAFR is a highly accurate, AI-based facial recognition platform architected to economically scale at high performance with rapid processing to detect and match millions of faces in real time. Recognize every fan, VIP, employee, or guest and surface key insights for a better experience.

Like many parents in the United States, Rob Glaser has been thinking a lot lately about how to keep his kids from getting shot in school. Specifically, he's been thinking of what he can do that doesn't involve getting into a nasty and endless battle over what he calls "the g-word." It's not that Glaser opposes gun control. A steady Democratic donor, Glaser founded the online streaming giant RealNetworks back in the 1990s as a vehicle for broadcasting left-leaning political views. It's just that any conversation about curbing gun rights in America tends to lead more to gridlock and finger-pointing than it does to action.