United States

To the selection committee: We would like to propose delivering this session in two formats--a 20 minute oral presentation and a hands-on workshop. The workshop format could be completed in 1.5-2 hours, so we would be happy to tailor it to a time frame that works for the conference. Please let me know if you have any questions.


Agenda:

How do you know if your data is behaving normally? Aerospace organizations face an ever-growing volume of time series sensor data. Detecting unusual patterns can be the key to improving performance or preventing failures. This session will explore techniques for time series anomaly detection, including statistical and AI-based approaches. Whether your focus is design, testing, or operations, you’ll learn how to uncover insights in sensor data that lead to more reliable and efficient systems.

Highlights include:

-Identifying anomalies in historical time series data
-Extracting and engineering features from time series signals
-Designing, comparing, and evaluating AI-based anomaly detection algorithms
-Deploying anomaly detection methods in real-time or streaming environments

NAML 2026

Anomaly Detection Techniques for Time Series and Sensor Data

signal

predictive maintenance

time series

anomaly detection

machine learning

We are pleased to announce the Thirty-Eighth AAAI Conference on Artificial Intelligence (AAAI-24), which will be held in Vancouver, British Columbia at the Vancouver Convention Centre – West Building from 20-27 February, 2024.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-24 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

We expect for AAAI-24 to be an in-person conference – one author of all accepted papers will be expected to present work in person unless there are exceptional circumstances that prevent this.

In order to access the AAAI-24 event page you need to register [here](https://aaai.org/aaai-conference/registration/)

AAAI 2024

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. 

technical paper

### Welcome to the 10th annual workshop on Naval Applications of Machine Learning!

NAML is the premier annual event showcasing current machine learning research relevant to Naval applications. We welcome participation from the defense community, industry, and academia. The NAML workshop is an opportunity for researchers from Navy research and development centers and our partner organizations to exchange technical ideas and to come together as a team. 
Naval Information Warfare Center Pacific is partnering with AFCEA San Diego chapter to host the workshop.
#### On-Demand Access

* This is an on-demand site and all the recorded content from the workshop is available here. 
* Access to the recordings is exclusively available to registered participants.
* Each registrant has received an email notification, with instructions on how to access the content.

[![](https://assets.underline.io/markdown_image/1/image/60cd12aa376c191d3467ea2ee60944ea.png)](https://naml2026.org/)

The goals of NAML are:

* To spread awareness of current machine learning research relevant to Naval applications
* To connect machine learning researchers with experts in military requirements
* To build and strengthen collaborations within the defense research community

#### *"What we do as a Navy is a human business. Don’t pass up an opportunity to come together as a team. Don’t pass up leadership opportunities, and don’t outsource them to a PowerPoint. If you have an opportunity to come together as a team, to increase belonging, do it."*
Admiral John M. Richardson, 31st Chief of Naval Operations

This is an on-demand site. All recorded content from NAML 2026 is available here and access to it is exclusively available to registered participants.

**Please log in using the email address you registered with.** If you have any issues logging in, please contact us at <naml2026@underline.io>.

Let’s celebrate the past decade by investing in the next. 🚀

Please log in to explore this event.

This is an on-demand site. All recorded content from NAML 2026 is available here and access to it is exclusively available to registered participants.

**Explore the latest advancements in machine learning tailored to Naval applications 🔎**

With on-demand access to NAML 2026, you can watch all lectures & browse poster presentations at your convenience, while also engaging directly with presenters via the Underline platform - ask questions, exchange ideas, and stay connected.

**If you still haven’t registered, you can do so now.**
👇
**Take advantage of a special one-time offer: full access for just $99, with complimentary registration for students, government, and military employees.**

**Get access here: https://naml2026-ondemand.paperform.co/**

Please check your spam folder, or contact our support team at **<naml2026@underline.io>** if you registered, but still haven't received an access email.

Let’s celebrate the past decade by investing in the next. 🚀

Please register!

The 10th Annual Workshop on Naval Applications of Machine Learning (NAML 2026) was held in San Diego, California, March 2-5, 2026. NAML is a forum for interdisciplinary experts from government, industry, and academia to discuss results of ongoing research on theory and applications of machine learning to address some of the most challenging problems faced by our naval forces. 



This tutorial/demo provides an introduction to the fundamentals of quantum computing from the software perspective, starting by demystifying the core concepts that differentiate quantum machines from classical ones, including qubits, superposition, and entanglement. Attendees will learn about quantum gates and how they form quantum circuits, highlighting the inherent differences in quantum algorithm design and the probabilistic nature of measurement. We will conclude with an application of quantum computing toward Machine Learning (QML), classifying simple data using a Variational Quantum Circuit (VQC). This provides a tangible glimpse into how these quantum principles are currently being researched for future applications to accelerate AI tasks.

Quantum Computing Fundamentals and Intro to Quantum Machine Learning

The "Co-Design for Edge AI" project is focused on developing and demonstrating an end-to-end co-design flow for integrating novel machine learning (ML) accelerators into specialized edge hardware optimized for mission size, weight, and power (SWaP) constrained applications. The project establishes tools and methodologies for identifying and implementing accelerators by analyzing application code to extract and convert computationally expensive portions into hardware description languages like Verilog. These accelerators target ML workloads that are currently performance bottlenecks, offering significant energy savings and enhanced mission capabilities compared to existing solutions. The co-design flow guides domain experts in concurrently designing hardware and software components, reducing design times from years to months, and enabling the rapid deployment of energy-efficient, specialized hardware for mission-critical applications such as UAVs, radio communications, and space satellite computing.
The project collaborates with various stakeholders to apply its co-design flow methodology across multiple application domains and partners. This approach is being transitioned to government and industry stakeholders, enabling early and informed hardware decisions. The project delivers application-specific accelerator designs that are significantly more energy-efficient than existing solutions, improving mission metrics by 20% and engaging multiple Department of Defense branches in the adoption and implementation of these advanced technologies.

Co-design for Edge AI

From biomedical sensing and monitoring to DVS camera output classification and RF fingerprinting, many applications require the ability to learn and adapt to spatiotemporal patterns, that is, patterns in multichannel time sequences. Spiking neural networks (SNNs) are bio-plausible computing models with high energy efficiency. The temporal dynamics of neurons and synapses enable them to detect temporal patterns and generate sequences. While Backpropagation Through Time (BPTT) is traditionally used to train SNNs, it is not suitable for online learning of embedded applications due to its high computation and memory cost as well as extended latency. In this work, we present Spatiotemporal Online Learning for Synaptic Adaptation (SOLSA), which is specifically designed for online learning of SNNs composed of Leaky Integrate and Fire (LIF) neurons with exponentially decayed synapses and soft reset. The algorithm not only learns the synaptic weight but also adapts the temporal filters associated to the synapses. Compared to the BPTT algorithm, SOLSA has much lower memory requirement and achieves a more balanced temporal workload distribution. Moreover, SOLSA incorporates enhancement techniques such as scheduled weight update, early stop training and adaptive synapse filter, which speed up the convergence and enhance the learning performance. When compared to other non-BPTT based SNN learning, SOLSA demonstrates an average learning accuracy improvement of 14.2%. Furthermore, compared to BPTT, SOLSA achieves a 5% higher average learning accuracy with a 72% reduction in memory cost. The extremely low computational cost of the learning algorithm allows us to perform both inference and training on the same FPGA platform, enabling true autonomy for mobile systems.

Neuromorphic Spatiotemporal Online Learning for Synaptic Adaptation

Naval operations increasingly depend on artificial intelligence (AI) models that process massive multi-modal sensor data to enable real-time perception, classification, and autonomous decision-making across ships, submarines, distributed sensors, and unmanned surface or underwater vehicles. These platforms operate in communication-denied, power-constrained, and harsh environments where cloud offloading and GPUs maybe impractical. Many naval assets, especially those at high altitude, in space-adjacent environments or near nuclear propulsion, also face elevated radiation exposure, making radiation-tolerant hardware essential. Conventional digital technologies incur significant overhead to mitigate radiation-induced faults, whereas emerging nonvolatile memories such as Resistive Random Access Memories (ReRAM) offer inherently higher radiation tolerance. In-memory computing (IMC) architectures further reduce energy and latency by eliminating the von Neumann data-movement bottleneck.
To meet these demands, we propose Waikiki, a ReRAM-based, radiation-tolerant edge AI accelerator designed for fast and efficient inference of deep neural networks (DNNs). Existing IMC accelerators struggle with modern AI models that require both linear layers and scaled dot-product attention, which rely on data-dependent matrix multiplications and complex nonlinear functions. They also require large, power-hungry analog-to-digital converters (ADCs) and spend most of their energy on analog-digital data processing. Further, they rely on external digital vector functional units for complex activations, limiting performance and suffer from low utilization when mapping full models to finite-size analog cores. Waikiki overcomes these limitations through the NL-DPE core, a Non-Linear Dot Product Engine that integrates conventional crosspoint arrays with ReRAM-based Analog Content Addressable Memories. This enables arbitrary nonlinear functions and data-dependent matrix multiplies to be executed directly in the analog domain as decision trees, while eliminating ADCs and enabling scalable radiation-robust analog computation. Noise resilience is further improved through software-based Noise Aware Fine-Tuning, avoiding costly in-device calibration and mitigating analog non-idealities.
We evaluate NL-DPE on diverse AI workloads, including CIFAR-10 and ImageNet, which are representative of embedded vision and onboard data-analysis tasks in naval environments. NL-DPE delivers 28× energy efficiency and 249× speedup over a GPU baseline at equal quantization, and 22× energy efficiency with 245× speedup compared to existing IMC accelerators, while maintaining high computational accuracy. Within Waikiki’s compact 64-core architecture, we demonstrate efficient model compression and mapping via singular value decomposition. The system achieves ResNet-50 inference at 7.5× compression, maintains >92% CIFAR-10 accuracy, and provides >3000× energy-efficiency improvement over the digital baseline.

Waikiki, an analog in-memory AI accelerator for edge environments

Illegal, unreported, and unregulated (IUU) fishing undermines maritime security and threatens the economic/political stability of coastal nations. For the U.S. Navy (hereafter, Navy) and allied forces, such activities represent not only an environmental or economic concern, but also a persistent intelligence and domain-awareness challenge that blurs the line between civilian and adversarial maritime behavior. Detecting, interpreting, and prioritizing these activities at scale requires automated systems that are both operationally reliable and analytically transparent.

This work presents an explainable, multi-stage machine learning methodology for large-scale detection and characterization of fishing activity using Automatic Identification System (AIS) telemetry provided by Global Fishing Watch (GFW). The system processes millions of vessel pings collected across varied ocean regions and time blocks to classify vessels as fishing or non-fishing, and—when applicable—to infer the likely fishing type (e.g., trawler, drifting longlines, purse seines, etc.). The model implements an XGBoost-based model template that processes various time-based statistics of AIS pings (e.g., instantaneous, daily, weekly, and historical AIS aggregations).

By using automated methods to identify fishing vessels based on their characterization, Navy operators can focus their attention on identifying vessels that are likely to be non-fishing. The examination of explainable results is key to making the best use of this system. This method integrates explainable AI (XAI) methods, including SHAP-based feature attribution, to help operators interpret model outputs and understand why vessels are identified/classified as they are. However, in isolation, these XAI methods have the potential to reduce operator efficiency, adding yet another set of data to interpret from the model. To mitigate this, we investigated the use of large language models (LLMs) to distill the model classifications, important features, and other XAI characteristics into a high-level, glanceable summary. These summaries enable operators to quickly assess the credibility of automated detections by highlighting key behavioral indicators such as motion irregularities, timing patterns, and spatial anomalies.

A complementary visualization interface, built on Folium, exposes these data and summaries, allowing analysts to rapidly explore regional patterns, identify anomalous behavior, and generate evidence-based assessments suitable for operational use. Ongoing evaluation is exploring the efficacy of the methods. This research contributes to the development of explainable, domain-aware machine learning pipelines for maritime situational awareness. Operationally, it provides investigations into a scalable, interpretable foundation for integrating AI-driven anomaly detection into Navy decision-support systems, advancing the broader goal of achieving efficient, trusted autonomy in maritime intelligence operations.

From Telemetry to Intelligence: An Explainable Pipeline for Fishing Activity Characterization

Spectrum AI Agent for Monitoring a Radio Dynamic Zone
Silvia Ionescu (sionescu@mitre.org)
Curtis Watson (cmwatson@mitre.org)

The increasing reliance on wireless communications highlights the critical need for innovative spectrum sharing strategies to maximize the utilization of finite resources. Dynamic spectrum environments require continuous monitoring and agile decision-making, yet the volume and complexity of radio frequency (RF) data can overwhelm human operator and hinder timely, accurate responses. There is a pressing need for tools that reduce cognitive overload while enabling robust, real-time spectrum coexistence in diverse, evolving operational contexts. 
The NSF National Radio Dynamic Zones (NRDZ) Spectrum Applications with Real-world Knowledge through Integrated Experiments (SPARKIE) project advances spectrum management through long-term dynamic spectrum sharing experimentation and the integration of cutting-edge artificial intelligence (AI) technologies. SPARKIE introduces a spectrum-aware AI agent that leveraged machine learning (ML) to continuously analyze RF data for dynamic, efficient monitoring and decision-making. By deploying sensing capabilities at edge sensor nodes equipped with high-performance GPUs, such as DeepWave AIR-T devices that combine software defined radio functionality with GPU acceleration, the project strives to enable real-time sensing, monitoring, and decision support. These edge-based tools are further augmented with large language model (LLM)-based agents that provide contextual reasoning and decision-making, spectrum monitoring, and evaluation of experimental outcomes. 
In this presentation, we will present results from the NRDZ AI spectrum agent’s evaluation of the November 2025 Hat Creek Radio Observatory spectrum coexistence experiments. Our results highlight the AI agent’s ability to assess experiment outcomes and deliver refinement suggestions that accelerate and improve decision-making. 

Approved for Public Release; Distribution Unlimited Public Release Case Number 25-02739-2. © 2025 The MITRE Corporation. ALL RIGHTS RESERVED


Spectrum AI Agent for Monitoring a Radio Dynamic Zone

The ability to predict catastrophic system events is important for computers and autonomous systems overall, in order to employ mitigations or at least minimize damage. Predicting these “tipping points” using machine learning is one way to forecast when a system will destabilize and the time before expected failure. Frequently, the response time between detection and implementing a mitigation or shutdown needs to be at machine speeds vs. human-response speeds.
Remote missions and environments would by necessity need to leverage greater autonomy in their systems and respective preventive measures. Adversaries may initiate a non-conventional cyber-attack, assumed by the attacker to be undetectable until it is too late. Similarly, complex production / processing systems may be on the verge of failure for multiple reasons without the obvious clues available in time for a human operator to take action. This goes beyond normal cyber mitigations and respective system predictive maintenance, as there is a clear need to forecast such failures.
A JHUAPL – funded research effort called “Tip-it” has started to develop and test specific classes of algorithms to forecast critical events. These algorithms adapt to the system by internally constructing nonlinear mathematical models of the system and respective environmental dependencies. This allows them to predict behaviors several time steps before they occur. The first year of this research is concluding with the following results:
a) Successfully forecasted impending destabilizations for both small and large systems.
b) Worked on multiple scales ranging from a simple physical, to power grid-sized systems.
c) Leveraged theoretical (nonlinear mathematical equations), real physical systems, actual static production plant data, and simulated power grid data. 
d) Learned / forecasted simple physical systems chaotic orbital with zero-shot learning. This lightweight ML algorithm class successfully learned the dynamic chaotic orbital at first exposure- predicting both onset and length of chaotic orbitals. 
These results and associated technical details were presented in May 2024 to NASA LaRC, who has since expressed interest in the progress of this research.
The follow-on focus would be on a re-stabilization hypothesis based upon earlier research, which indicated that a few preselected perturbations can be added to a theoretical model in order to extend its stability beyond its normal performance envelope.
The end-objective would be twofold: a dynamically re-stabilizing system that is more resilient to attacks or failures, and a means of discretely de-stabilizing an adversary’s system.


Predicting Destabilizations in Complex Systems and Mitigations that Extend Performance

Underwater acoustic vector sensor (AVS) packages measure both acoustic pressure and particle velocity, enabling them to estimate bearings and elevations of acoustic signals without significant spatial aperture. AVS sensors play a crucial role in sonobuoy, towed array, and autonomous applications. While beamforming techniques are commonly applied to these sensor outputs, a range of additional nonlinear metrics can be derived from AVS using underlying physical principles. Such metrics include "azigrams," "elegrams," the kinetic-to-potential-energy ratio (K/E), polarization properties, and the normalized transport velocity (NTV), a measure that embeds information about the angular spread of an acoustic source. These metrics provide alternative time-frequency representations of data beyond a simple spectrogram and can be arranged as channels fed into an autoencoder (AE) for unsupervised learning and feature space clustering, similar to how camera images feed separate red, green, and blue channels into convolutional layers of an AE. We explore and demonstrate these concepts using hundreds of thousands of bowhead whale sounds collected in arctic waters between 2008 and 2014 and examine whether K/E and NTV metrics improve AE discrimination of angularly compact acoustic signals from both non-acoustic transients (e.g., knocks and vibrations) and transient signals from distributed sources (e.g., wind and reverberation).

Applying physics-based insights to unsupervised learning on acoustic vector sensors

The threat of quantum computers breaking widely used public-key cryptography like RSA and ECC grows steadily as the machines mature. Fortunately, NIST has standardized a set of cryptographic algorithms known as Post-Quantum Cryptography (PQC) that are expected to be secure against both classical and quantum attacks. Migrating to PQC as soon as possible is critical for the Navy for several reasons:
1. Long-term security needs: The Navy's communications are often confidential and frequently require decades-long security. Communication integrity is essential for fleet operations and mission success. Post-quantum cryptography will protect this data against future quantum computing threats.
2. Harvest Now, Decrypt Later (HNDL): Nation-states are collecting encrypted communications now in anticipation of the day when a future quantum computer will be able to decrypt them. Regardless of when this occurs, the HNDL strategy provides urgency for the Navy to move away from vulnerable cryptography soon.
3. Infrastructure longevity: Naval systems and vessels have long operational lifespans (20+ years). Implementing quantum-resistant cryptography now will protect these long-lived assets against future threats.
4. Supply chain security: The Navy relies on complex supply chains that must be secured against potential quantum computing attacks.

However, there are multiple challenges in updating the Navy’s computing and communications systems. First, all vulnerable cryptography needs to be inventoried, whether it exists across the wire or on a network endpoint. Second, these unsafe algorithms need to be replaced with PQC while maintaining some minimum level of network performance. Finally, the migration should also build crypto-agility into the system so that future cryptographic transitions can be performed efficiently.

Noblis is addressing these challenges through our internal research. We have met with several vendors that offer cryptographic inventorying tools and have down-selected to the three best tools. We are also building a virtual network testbed that allows us to test and evaluate these tools to select the best one for our clients. We will also evaluate the new PQC algorithms on this testbed to understand and mitigate the effects of longer key lengths on network performance. Finally, we will use the testbed to develop strategies for improving crypto-agility.

We will be exploring the use of AI/ML during each step of internal research process, including for creating a representative set of networks for quickly evaluating inventorying tools and for identifying locations in a large network where vulnerable cryptography or high priority datasets are most likely to be present. By transitioning to PQC in a timely manner, the Navy can maintain communications security and operational effectiveness in an era when quantum computing will ultimately render current encryption methods unsafe.


Migrating to Post-Quantum Cryptography

Achieving optimal performance from machine learning (ML) algorithms in complex naval applications requires effective workload management.
This effort presents a novel ML Control network designed to enhance signal processing in challenging environments. The network intelligently routes data to specialized functions tailored for specific data characteristics, enabling optimized baseband extraction. We are currently investigating integrating this architecture into a Low Band Universal Communications System Receiver for submarine platforms, addressing the unique operational requirements of very low frequency (VLF) communications. The objective is to improve both bit error rate and anti-jamming performance while minimizing computational burden – a critical need for resource-constrained naval systems. This work seeks to streamline processing architectures and improve size, weight, power, and cost (SWAP-C) metrics, ultimately contributing to enhanced receiver designs for Naval Command, Control, and Communications (NC3) capabilities. We believe this research offers a promising approach to improving ML performance in demanding naval scenarios and invites collaboration with other researchers exploring similar challenges within the defense research community.


Premium content

Downloads

Next from NAML 2026

Quantum Computing Fundamentals and Intro to Quantum Machine Learning

Similar lecture

When Model Meets New Normals: Test-Time Adaptation for Unsupervised Time-Series Anomaly Detection