Daily Tech Digest - May 14, 2024

Transforming 6G experience powered by AI/ML

While speed has been the driving force behind previous generations, 6G redefines the game. Yes, it will be incredibly fast, but raw bandwidth is just one piece of the puzzle. 6G aims for seamless and consistent connectivity everywhere. ... This will bridge the digital divide and empower remote areas to participate fully in the digital age. 6G networks will be intelligent entities, leveraging AI and ML algorithms to become: Adaptive: The network will constantly analyze traffic patterns, user demands, and even environmental factors. Based on this real-time data, it will autonomously adjust configurations, optimize resource allocation, and predict user needs for a truly proactive experience. Imagine a network that anticipates your VR gaming session and seamlessly allocates the necessary resources before you even put on the headset. Application-Aware: Gone are the days of one-size-fits-all connectivity. 6G will cater to a diverse range of applications, each with distinct requirements. The network will intelligently recognize the type of traffic – a high-resolution video stream, a critical IoT sensor reading, or a real-time AR overlay – and prioritize resources accordingly. This ensures flawless performance for all users, regardless of their activity.

How data centers can simultaneously enable AI growth and ESG progress

Unlocking AI’s full potential may require organizations to make significant concessions on their ESG goals unless the industry drastically reduces AI’s environmental footprint. This means all data center operators - including both in-house teams and third-party partners - must adopt innovative data center cooling capabilities that can simultaneously improve energy efficiency and reduce carbon emissions. The need for HPC capabilities is not unique to AI. Grid computing, clustering, and large-scale data processing are among the technologies that depend on HPC to facilitate distributed workloads, coordinate complex tasks, and handle immense amounts of data across multiple systems. However, with the rapid rise of AI, the demand for HPC resources has surged, intensifying the need for advanced infrastructure, energy efficiency, and sustainable solutions to manage the associated power and cooling requirements. In particular, the large graphics processing units (GPUs) required to support complex AI models and deep learning algorithms generate more heat than traditional CPUs, creating new challenges for data center design and operation. 

Cutting the cord: Can Air-Gapping protect your data?

The first challenge is keeping systems up to date. Software requires patching and upgrading as bugs are found and new features needed. An Air-Gapped system can be updated via USB sticks and CD-Roms, but this is (a) time consuming and (b) introduces a partial connection with the outside world. Chris Hauk, Consumer Privacy Advocate at Pixel Privacy, has observed the havoc this can cause. “Yes, hardware and software both can be easily patched just like we did back in the day, before the internet,” says Hauk. “Patches can be ‘sneakernetted’ to machines on a USB stick. Unfortunately, USB sticks can be infected by malware if the stick used to update systems was created on a networked computer. “The Stuxnet worm, which did damage to Iran’s nuclear program and believed to have been created by the United States and Israel, was malware that targeted Air-Gapped systems, so no system that requires updating is absolutely safe from attacks, even if they are Air-Gapped.” The Air-Gap may suffer breaches. Users may want to take data home or have another reason to access systems. A temporary connection to the outside world, even via a USB stick, poses a serious risk.

Delivering Software Securely: Techniques for Building a Resilient and Secure Code Pipeline

Resilience in a pipeline embodies the system's ability to deal with unexpected events such as network latency, system failures, and resource limitations without causing interruptions. The aim is to design a pipeline that not only provides strength but also maintains self-healing and service continuity. By doing this, you can ensure that the development and deployment of applications can withstand the inevitable failures of any technical environment. ... To introduce fault tolerance into your pipeline, you have to diversify resources and automate recovery processes. ... When it comes to disaster recovery, it is crucial to have a well-organized plan that covers the procedures for data backup, resource provision, and restoration operations. This could include automating backups and using CloudFormation scripts to provision the infrastructure needed quickly. ... How can we ensure that these resilience strategies are not only theoretically effective but also practically effective? Through careful testing and validation. Use chaos engineering principles by intentionally introducing defects into the system to ensure that the pipeline responds as planned. 

Cinterion IoT Cellular Modules Vulnerable to SMS Compromise

Cinterion cellular modems are used across a number of industrial IoT environments, including in the manufacturing and healthcare as well as financial services and telecommunications sectors. Telit Cinterion couldn't be immediately reached for comment about the status of its patching efforts or mitigation advice. Fixing the flaws would require the manufacturer of any specific device that includes a vulnerable Cinterion module to release a patch. Some devices, such as insulin monitors in hospitals or the programmable logic controllers and supervisory control and data acquisition systems used in industrial environments, might first need to be recertified with regulators before device manufacturers can push patches to users. The vulnerabilities pose a supply chain security risk, said Evgeny Goncharov, head of Kaspersky's ICS CERT. "Since the modems are typically integrated in a matryoshka-style within other solutions, with products from one vendor stacked atop those from another, compiling a list of affected end products is challenging," he said. 

Automotive Radar Testing and Big Data: Safeguarding the Future of Driving

In radar EOL testing, one of the key verification parameters is the radar cross-section (RCS) detection accuracy, which represents the size of an object. Unlike passive objects that have fixed RCS, RTS allows the simulation of various levels of RCS, echoing a desired object size for radar detection. While RTS systems offer versatility for radar testing, they present challenges to overcome. One such challenge is the sensitivity of the system’s millimeter-wave (mmWave) components to temperature variations, which can significantly impact the ability to accurately simulate RCS values. Therefore, controlling the ambient temperature in a testing setup is important to ensuring that the RTS replicates the RCS expected for a given object size. Furthermore, the repercussions extend beyond the immediate operational setbacks with. the need to scrap a number of radar faulty module units. Not only does this represent a direct monetary loss and the overall profit margin, but it also contributes to waste and environmental concerns. All these adverse outcomes, from reduced output capacity to financial losses and environmental impact, highlight the critical importance of integrating analytics software into an automotive radar EOL testing solution. 

Nvidia teases quantum accelerated supercomputers

The company revealed that sites in Germany, Japan, and Poland will use the platform to power quantum processing units (QPU) in their high performance computing systems. “Quantum accelerated supercomputing, in which quantum processors are integrated into accelerated supercomputers, represents a tremendous opportunity to solve scientific challenges that may otherwise be out of reach,” said Tim Costa, director, Quantum and HPC at Nvidia. “But there are a number of challenges between us, today, and useful quantum accelerated supercomputing. Today’s qubits are noisy and error prone. Integration with HPC systems remains unaddressed. Error correction algorithms and infrastructure need to be developed. And algorithms with exponential speed up actually need to be invented, among many other challenges.” ... “But another open frontier in quantum remains,” Costa said. “And that’s the deployment of quantum accelerated supercomputers – accelerated supercomputers that integrate a quantum processor to perform certain tasks that are best suited to quantum in collaboration with and supported by AI supercomputing. We’re really excited to announce today the world’s first quantum accelerated supercomputers.”

Tailoring responsible AI: Defining ethical guidelines for industry-specific use

As AI becomes increasingly embedded in business operations, organizations must ask themselves how to prepare for and prevent AI-related failures, such as AI-powered data breaches. AI tools are enabling hackers to develop highly effective social engineering attacks. Right now, having a strong foundation in place to protect customer data is a good place to start. Ensuring third-party AI model providers don’t use your customers’ data also adds protection and control. There are also opportunities for AI to help strengthen crisis management. The first relates to security crises, such as outages and failures, where AI can identify the root of an issue faster. AI can quickly sift through a ton of data to find the “needle in the haystack” that points to the source of the attack or the service that failed. It can also surface relevant data for you much faster using conversational prompts. In the future, an analyst might be able to ask an AI chatbot that’s embedded in its security framework questions about suspicious activity, such as, “What can you tell me about where this traffic originated from?” Or, “What kind of host was this on?”

Taking a ‘Machine-First’ Approach to Identity Management

With microservices, machine identities are proliferating at an alarming rate. Cyberark has reported that the ratio of machine identities to humans in organizations is 45 to 1. At the same time, 87% of respondents in its survey said they store secrets in multiple places across DevOps environments. Curity’s Michal Trojanowski previously wrote about the complex mesh of services comprising an API, adding that securing them is not just about authenticating the user. “A service that receives a request should validate the origin of the request. It should verify the external application that originally sent the request and use an allowlist of callers. ... Using agentless scanning of the identity repositories engineers are using and log analysis, the company first maps all the non-human identities throughout the infrastructure — Kubernetes, databases, applications, workloads, and servers. It creates what it calls attribution— a strong context of which workloads and which humans use each identity, including an understanding its dependencies. Mapping ownership of the various identities also is key. “Think about organizations that have thousands of developers. Security teams sometimes find issues but don’t know how to solve them because they don’t know who to talk with,” Apelblat said.

The limitations of model fine-tuning and RAG

Several factors limit what LLMs can learn via RAG. The first factor is the token allowance. With the undergrads, I could introduce only so much new information into a timed exam without overwhelming them. Similarly, LLMs tend to have a limit, generally between 4k and 32k tokens per prompt, which limits how much an LLM can learn on the fly. The cost of invoking an LLM is also based on the number of tokens, so being economical with the token budget is important to control the cost. The second limiting factor is the order in which RAG examples are presented to the LLM. The earlier a concept is introduced in the example, the more attention the LLM pays to it in general. While a system could reorder retrieval augmentation prompts automatically, token limits would still apply, potentially forcing the system to cut or downplay important facts. To address that risk, we could prompt the LLM with information ordered in three or four different ways to see if the response is consistent. ... The third challenge is to execute retrieval augmentation such that it doesn’t diminish the user experience. If an application is latency sensitive, RAG tends to make latency worse. 

Quote for the day:

"What you do makes a difference, and you have to decide what kind of difference you want to make." -- Jane Goodall

No comments:

Post a Comment