Digital Twin for Vulnerability Assessments

I’m thrilled to share that I’ll present best practices to deterministic, ultra-low-latency inference for ML-based trading systems at STAC Research upcoming event in New York City on October 28th. In my session, I’ll cover several key points: - How model structure impacts determinism and latency. - The role of quantization in balancing accuracy, memory footprint and latency. - How different deployment targets, from CPUs to specialised accelerators, affect latency. - Real-world benchmarks demonstrating how these system-level design choices enable ML strategies that can reliably operate at HFT speeds. This session is particularly relevant for those working at the intersection of ML, systems design, and trading.  Always happy to connect and exchange ideas! #STACResearch #LowLatency #HighFrequencyTrading #MachineLearning #InferenceOptimization #Quantization #FinTech

The move to high-capacity SDDs is less a storage upgrade than a structural shift in how data infrastructure is designed for the AI era. Read more about the real-world challenges HDDs present, and how SSDs may hold the key to unlocking efficiency at scale.

Scaling autonomous agents requires more than just advanced models—it requires trust. That’s why LazAI is partnering with Zypher Network, a team pioneering prompt integrity verification and proof-of-prompt security at the protocol level. This collaboration ensures that agents built and deployed on LazAI are safer, more transparent, and exhibit verifiable behavior—a critical step toward enterprise-grade AI systems you can rely on.

Computer vision has long been limited by fragmentation. Siloed models, devices, and frameworks; typically optimized for a single use case. Visual General Intelligence (VGI) shares learning across contexts: i.e., a model trained to detect hazards can inform a model monitoring process efficiency, because it sees the same environment from different angles. That cross-domain learning is how VGI achieves scale, becoming a unifying layer across operations to connect safety, quality, and performance data into a single perceptual system. Our VGI whitepaper explores how enterprises can architect this today using existing infrastructure, and how visual intelligence will evolve from isolated deployments into connected ecosystems. Read it here 🔗 https://lnkd.in/dHPn2UVN

🚀 Excited to share that my latest paper, "Observer-based Adaptive Fault-Tolerant Event-Triggered Consensus for Linear Multiagent Systems: A Chattering-Free Approach" has been published in the International Journal of Systems Science (SCI, Q1, IF ≈ 4.6)! In this work, we tackle the distributed observer-based adaptive event-triggered consensus problem for linear multi-agent systems with time-varying multiplicative actuator faults and bounded uncertainties, a challenging scenario for robust coordination. Key highlights: Developed an adaptive event-triggered distributed observer to estimate unavailable state information, avoiding reliance on actual system states. Co-designed a real-time fault estimation mechanism with adaptive coupling gains, enhancing fault tolerance without depending on global communication network parameters and ensuring scalability. Proposed a nonlinear compensation term with a time-varying boundary layer width, improving robustness against non-identical bounded actuator uncertainties while eliminating chattering. Introduced an adaptive dynamic event-triggered mechanism (ADETM) incorporating fault estimation and adaptive coupling weights to ensure robustness against time-varying multiplicative actuator faults while reducing communication overhead. The approach guarantees asymptotic leader-following consensus, rules out Zeno behavior, and reduces unnecessary event triggers for efficient communication. Simulation results demonstrate its effectiveness in enabling communication-efficient, robust, and coordinated multi-agent systems. Read the full paper here: https://lnkd.in/gmv9369F #ControlSystems #MultiAgentSystems #EventTriggeredControl #FaultTolerance #AdaptiveControl #ResearchPublication #SystemsScience

VibroAI’s automated vibration analysis system converts raw sensor data into actionable maintenance insights through a structured series of stages. Each step builds on the previous one, creating a comprehensive diagnostic framework that unites proven signal processing methods with advanced artificial intelligence.

𝑸𝒖𝒆𝒔𝒕𝒊𝒐𝒏: Frequency hopping was once the gold standard for jamming resistance, jumping across channels to avoid detection faster than the enemy could react. But in modern electronic warfare, jammers aren’t just louder, they’re smarter. - What happens when a jammer doesn’t need to sweep the whole band but instead uses real-time sensing to predict your next hop? - Can fast hopping still guarantee communication reliability in the presence of a smart jammer? - When does frequency agility stop being enough and how do we calculate the tipping point where evasion fails? 𝑨𝒏𝒔𝒘𝒆𝒓: In FHSS systems, communication is protected by spending only a short time per frequency called the dwell time before hopping to the next. But cognitive jammers now use fast spectrum sensing and machine learning to predict or react within microseconds. To evaluate how vulnerable a system is, we compare: → t_dwell = how long we stay on a single frequency → t_react = how quickly the jammer detects and tunes to that frequency We can calculate the probability of jamming success using: → P_jam_success = 1 − e^(−t_dwell / t_react) The longer you dwell, the more likely the jammer can catch you. If your system is hopping fast but the jammer is reacting even faster, the entire link becomes vulnerable. Let’s say we have a tactical radio using FHSS with the following specs: → Hop bandwidth: 100 MHz → Hop rate: every 75 µs → Jammer reaction time: 20 µs (FFT detection + tuning delay) We calculate: → P_jam_success = 1 − e^(−75 / 20) → P_jam_success ≈ 1 − e^(−3.75) ≈ 0.976 That’s a 97.6% chance of being jammed on every hop despite hopping over 13,000 times per second. Even worse, if the jammer predicts hop timing, it doesn’t even need to react, it simply preloads the interference into the right slot. The result isn’t just packet loss, it’s full communication collapse. Modern jammers can: - Detect signals across wide bandwidths in real time - Deinterleave hopping patterns with machine learning - Launch targeted bursts that line up with known slot timing To survive in this environment, FHSS systems must evolve which means: - Injecting timing jitter to disrupt predictability - Adding spread spectrum redundancy - Using spatial filtering (beamforming) to nullify jammers - Designing for unpredictable behavior and not just fast behavior The image below shows how Frequency Hopping Spread Spectrum (FHSS) works, the transmitter switches rapidly between multiple carrier frequencies (f0–f7) following a pseudorandom sequence. Each channel carries equal power and the hopping pattern over time ensures the signal doesn’t linger on one frequency long enough for a jammer to track or block it, illustrating how FHSS enhances resilience against interference. #ElectronicWarfare #SmartJamming #FHSS #DSP #CommsJamming #SpectrumWarfare #WirelessSecurity #SignalProcessing #RFEngineering #AntiJam #MilitaryTech #PhDResearch #TacticalComms #JammingResistance

Exaforce: Experience-led multi-model AI for SOC resolution - Exaforce distills telemetry signals from a wide variety of security, observability, and software delivery tool chains into a semantic funnel that assigns enterprise environment knowledge and behavioral threat context to security operations work. https://lnkd.in/eTPSU-n4

Government’s tech department has recently conducted specialised tests designed to determine the extent to which artificial intelligence models could launch and coordinate automated cyberattacks. Story ⬇️

Government’s tech department has recently conducted specialised tests designed to determine the extent to which artificial intelligence models could launch and coordinate automated cyberattacks. Story ⬇️