United States

Large Reasoning Models (LRMs) demonstrate strong performance on complex reasoning tasks, yet they often suffer from overthinking—producing excessively long chain-of-thought (CoT) traces that increase inference cost and may degrade accuracy. Our analysis reveals a clear anti-correlation between reasoning length and accuracy, where across multiple stochastic decodes, the short reasoning paths consistently achieve the highest correctness, while longer ones accumulate errors and repetitions. These short optimal reasoning paths can be found ideally through full enumeration of the reasoning space. However, the tree-structured reasoning space grows exponentially with sequence length, rendering exhaustive exploration infeasible. To address this, we propose DTS, a model-agnostic decoding framework that sketches the reasoning space by selectively branching at high-entropy tokens and applies early stopping to select the shortest completed reasoning path. This approach approximates the optimal solution that enhances both efficiency and accuracy, without requiring additional training or supervision. Experiments on AIME2024 and AIME2025 datasets with DeepSeek-R1-Distill-Qwen-7B and 1.5B show that DTS improves accuracy by up to 8%, reduces average reasoning length by 23%, and decreases repetition frequency by 12%, demonstrating DTS’s ability for scalable and efficient LRM reasoning.

NAML 2026

DTS: Enhancing Large Reasoning Models via Decoding Tree Sketching

efficient reasoning

machine learning

large language models

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. 
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) 
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

**Online Registration Form**: https://acl.swoogo.com/acl2026

Registration Required

ACL 2026

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

poster

### 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. 



Subject: 

Creation of a Multidimensional Multimodal Anomaly Detection with Interpretability (MADI) model using in-flight sensor data to augment fault prediction and fleet health monitoring predictive maintenance modeling efforts for the F-15E aircraft. 



Abstract:  

This work presents an end-to-end MADI prototype developed to enhance anomaly detection capabilities using F-15E Flight Data. Flight data was ingested and transformed within a specialized preprocessing pipeline designed to stabilize raw flight data, align disparate recordings, and support a repeatable anomaly detection framework. To address the absence of labeled fault data in operational settings, synthetic anomalies were generated using a uniform negative-sampling procedure, enabling supervised model training without dependence on historical failure events. 

A custom deep neural network was trained and tuned using a reproducible MLOps framework with hyperparameter optimization and automated experiment tracking. The model produces per-timestamp anomaly probabilities and applies integrated-gradients attribution to identify which sensor channels most strongly contribute to each detected anomaly. This approach treats the ensemble state of all sensors simultaneously and enables detection of complex feature-cross anomalies missed by univariate and rule-based algorithms. These outputs are aggregated into sortie-level metrics that summarize anomaly prevalence, severity, and subsystem-level contribution patterns. Results are integrated into a Historical Maintenance Timeline to support alignment with flight activity, maintenance actions, and readiness events. This prototype demonstrates a scalable, explainable, and operationally relevant anomaly detection capability for fighter aircraft sensor data. The framework supports DoD objectives in condition-based maintenance, fleet health monitoring, and data-driven sustainment planning, and establishes a foundation for future enhancements including anomaly-sequence contraction, event correlation, subsystem mapping, and operational-mode-aware anomaly models.

MADI-Based Anomaly Detection to Enhance Predictive Maintenance for the F-15E

The Results-Aligned Performance Improvement & Development (RAPID) training system transforms training content creation by seamlessly blending artificial intelligence with human expertise. This innovative solution enables Naval commands to quickly create and update training materials when operational procedures and Naval policies change, ensuring fleet readiness. Our presentation will demonstrate a clickable prototype showcasing how RAPID can be deployed for critical naval training needs such as shipboard damage control procedures, tactical operations, and maintenance protocols where accuracy and compliance with Navy standards are paramount. Attendees will observe how instructional systems designers (ISDs) configure parameters, upload Navy-specific documentation, and generate draft content that both ISDs and subject matter experts (SMEs) then review and refine. RAPID stands apart from competitors in five key ways: 1) Unlike commercial AI solutions requiring cloud connectivity, RAPID can be rapidly deployed within secure networks, with no external dependencies, making it ideal for fleet-wide deployment. 2) While other systems focus on automating content creation, RAPID’s human-in-the-loop workflow preserves the ISD’s and SME’s expertise while ensuring critical human oversight. 3) The system’s parameter pack assistant allows users to customize and standardize the output logic and format of AI-generated content in compliance with Navy training standards, learning taxonomies, and domain-specific constraints across different learning contexts. 4) RAPID employs a specialized retrieval-augmented generation (RAG) architecture that can index Navy-specific documentation, ensuring all generated training content maintains fidelity to authorized naval sources and terminology. 5) The system's AI validation engine quantitatively measures content against customer-defined standards using proprietary algorithms for coverage (ensuring all objectives are addressed), groundedness (verifying factual accuracy), congruency (confirming alignment between objectives, content, and assessments), and measurability (ensuring outcomes can be properly evaluated).

RAPID Training System

Humanitarian Assistance and Disaster Relief (HADR) operations require rapid damage assessment to effectively allocate resources and save lives, yet traditional satellite imagery analysis can take days to complete. This delay creates a critical gap between disaster occurrence and informed response, costing lives and resources. Our team at CMU's Software Engineering Institute (SEI) presents a modernized, cloud-based platform that enables domain experts and mission users to deploy custom computer vision models for damage assessment within minutes rather than days, significantly accelerating response times across multiple Areas of Responsibility (AORs).
Our approach addresses a fundamental challenge in operational disaster response: the need to quickly adapt pre-trained models to new, out-of-distribution imagery from emerging incidents with diverse geographic, structural, and disaster characteristics. CMU SEI previously co-hosted the xView2 Prize Challenge and deployed machine learning solutions for EUCOM (Turkey & Ukraine, 2022-2024) and CENTCOM (Syria 2023). However, we observed critical limitations regarding fragmented workflows for model training and inferencing. Our platform leverages an image segmentation base model pre-trained on satellite imagery to predict damage classes across disaster types. When users encounter new scenarios, they upload imagery and rapidly train lightweight model adapters using LoRA (Low-Rank Adaptation), a Parameter-Efficient Fine-Tuning technique requiring limited computational resources and training data.
The adapter-based architecture provides several operational advantages: trained adapters can be reused across similar scenarios, fine-tuned for evolving situations, and shared across organizations. Domain experts can rapidly customize models for their specific AORs without deep machine learning expertise, democratizing access to advanced AI capabilities.
Our platform's cloud-agnostic architecture enables deployment on both commercial and government cloud environments and on-premises infrastructure, operating seamlessly from tactical edge environments to strategic operations centers, enabling cross-echelon collaboration. This infrastructure-agnostic design ensures damage assessment capabilities remain available regardless of network constraints or deployment environment, making it a force multiplier for operations where interoperability and rapid information sharing are paramount.
Preliminary results demonstrate that providing limited (2-16) labels to post-disaster satellite imagery yields substantial performance improvements over the base model, with low computational costs measured in minutes of GPU time. This efficiency is critical for first-wave response scenarios where every minute impacts mission outcomes. By enabling iterative refinement as new imagery arrives, our platform empowers Battle Damage Assessment to make data-driven resource allocation decisions at the speed required by modern HADR operations.

Computer Vision for Accelerating HADR

The Finding Responsive Intelligence in Multimodal Analysis (FRIDAY) project seeks to accelerate the development of artificial general intelligence (AGI) by integrating multiple, complex holonic algorithms into high‑fidelity flight simulators for next‑generation drone training and performance enhancement. Building on a one‑year deep‑analysis effort, FRIDAY proposes a hierarchical control architecture—Human → LLM Manager → LLM Executors—through which mission directives are translated into actionable drone behaviors. 

Central to the effort is an agentic LLM system that orchestrates swarms of specialized LLM‑agents, each tasked with mastering distinct operational capabilities. By grounding these agents in a modular skill set, the system can rapidly configure and deploy drone units tailored to diverse mission profiles. 

If realized, FRIDAY could transform military drone operations: autonomous platforms would dynamically adapt to battlefield conditions, conduct sophisticated reconnaissance, and execute precision tasks with minimal human oversight. The modular skill framework enables swift fielding of specialized drone formations for applications ranging from search‑and‑rescue and persistent surveillance to target acquisition and kinetic engagement. 

This work outlines the conceptual design, simulation‑based validation strategy, and anticipated operational benefits, offering a roadmap for leveraging AGI‑driven autonomy especially in an offline manner.

Finding Responsive Intelligence in Multimodal Analysis (FRIDAY) project

As Naval forces increasingly operate in contested, disconnected, intermittent, and limited environments, reliance on cloud-tethered Artificial Intelligence becomes a critical vulnerability. Large Language Models and Generative AI offer clear value for predictive maintenance, technical document synthesis, and decision support, but most current architectures depend on reach-back to hyper-scale data centers. This dependency is not viable for vessels operating under EMCON or in bandwidth-denied theaters.

This presentation examines the architectural and operational considerations for deploying sovereign AI at the tactical edge. The focus is the convergence of ruggedized compute and air-gapped Generative AI models running entirely within the shipboard perimeter.

We introduce a reference architecture for a disconnected maintenance advisor. The system uses Retrieval-Augmented Generation to ingest large volumes of unclassified and CUI technical manuals. By hosting the embedding models, vector database, and inference engine locally on edge hardware, operators can query complex repair procedures and receive context-aware citations without emitting traffic or relying on satellite links.

A key element is the methodology used to validate these systems before deployment. The discussion covers how edge AI stacks can be stressed against simulated tactical constraints and the technical challenges that emerge during integration, including:
Inference Optimization: Balancing model quantization and accuracy to fit efficient LLMs within tactical hardware limits.
Multimodal Ingestion: Using on-premise OCR and vision models to convert analog or handwritten maintenance logs into digital inputs.
Hallucination Control: Enforcing strict RAG patterns that require the model to cite specific technical manuals, reducing the risk of generative error in critical repair workflows.

This session focuses on the practical implementation of sovereign, self-contained AI capabilities within disconnected environments. By pairing edge-ready infrastructure with rigorous validation practices, Naval forces can reduce mean time to repair and maintain information advantage regardless of connectivity conditions.

Deploying Disconnected Generative AI for Naval Maintenance and Decision Superiority

In many areas of the world, changes in tactical threat capabilities have significantly increased the threat to U.S. and allied forces. Our ability to rapidly acquire, provide high confidence target identification, track, and estimate increasingly challenging, dynamic targets in high intensity environments using EO, RF, and EW sensors is critical to mission effectiveness.
Liquid AI is a new class of time-integrative neural networks from dynamical systems theory that are ideally suited to continuous-time data analysis. Unlike traditional feed-forward neural networks, Liquid AI learns complex spatiotemporal patterns using high lateral interconnectivity. Data from position encoders, velocity and accelerometers (found on UAVs), LADAR, 3D imaging (found in manufacturing processes), Dynamic Vision Sensor (DVS or event cameras), and time-varying signals in the audio to higher radar frequencies are aptly suitable for Liquid AI signal processing. 
Results from 2025 Raytheon Internal Research and Development (IRaD) on Liquid AI showed promising results open-source data. A solution was then developed using Raytheon-proprietary data. It was shown to outperform an industry standard (i.e., hand-crafted algorithms) as well as a current neural network solution, using 4-6X fewer parameters. In the literature, Liquid AI uses up to 62X fewer parameters, making it ideal for low-SWaP high throughput applications.
In this presentation, we present the different types of Liquid AI, including Liquid Time-Constant (LTC) networks, Neural ODE (Ordinary Differential Equations), and Continuous-time Recurrent Neural Networks (CTRNN). We then explain various machine learning approaches, such as backpropagation-of-errors, surrogate modeling & translation, and ODE hyper-parameter tuning. We outline our technical approach and show comparative results from the 2025 IRaD. Finally, we present a generalized roadmap showing what types of problems are applicable to novel Liquid AI solutions.


Spatiotemporal Processing with Liquid AI

This project explores lessons learned in designing, prototyping, and evaluating AI agents for intelligence analysis for a cyber defense use case. Despite growing interest, there remains little practical guidance on how to design and build AI agents, particularly guidance grounded in real-world implementations rather than vendor marketing. This project investigates the unique challenges of workflow design and agent evaluation, especially for novel or evolving processes.

Agentic AI systems including AI agents leverage augmented Large Language Models (LLMs) to autonomously complete tasks, workflows, or make decisions. Augmented LLMs are LLMs equipped with tools, access to memory, and retrieval capabilities. Using the capabilities of an augmented LLM, agents can autonomously search for information by generating their own queries, can select and use tools to complete specific actions, and can store and recall information to build context and improve performance.

Overall, AI agents have the potential to increase speed and scale, which may mean freeing up humans to handle more challenging or creative tasks. However, AI agents also present unique challenges, in that they are non-deterministic systems, meaning they can produce stochastic, and in some cases, incorrect, results. Furthermore, AI agents can rely on many Machine Learning (ML) components, increasing the chance of cascading errors.

This work presents lessons learned about the opportunities and challenges related to building AI agents. This work centers around a case study in which the team designed, prototyped, and evaluated two AI agents, built from scratch in LangGraph, for a cyber operations use case. The team worked closely with a Subject Matter Expert (SME) to design and refine the agents.

The first agent defines Priority Intelligence Requirements (PIRs) based on inputs such as the ACARE-I threat actors, a terrain map, and commanders' intent. The second agent then takes the same inputs, along with the newly created PIRs, to identify associated MITRE ATT&CK Tactics, Techniques, and Data Sources that a cyber team will need to be aware of for their upcoming mission.

Through this work, the team developed a set of practical recommendations for teams looking to build AI agents for mission critical use cases. These recommendations range from guidance on incorporating traceability into agents during early design, to implementing agents modularly throughout development and experimentation. These lessons learned are intended to help guide future agentic AI projects in developing agents iteratively with stakeholders, designing agents in a manner that is useful and user-centered, and identifying where traditional software engineering and ML practices are still required. This work presents actionable guidance informed by operator experience to support effective adoption by government organizations such as the Department of War (DoW).

Agentic AI Unpacked: Design and Engineering Lessons from a Mission-Focused Case Study

Reliable vehicle tracking from remote sensing data often requires robust state estimation 
across widely varying temporal resolutions. In this work, we propose a physics-informed 
attentive neural process (PI-AttNP) for multi-scale vehicle tracking that explicitly integrates 
simplified kinematic models as structural priors within a neural process framework. This 
design enforces adherence to physical motion laws while reducing error and accelerating 
convergence. The model leverages time-aware attention weights to capture statistical 
dependencies across irregularly spaced observations and employs split conformal 
prediction to provide distribution-free uncertainty guarantees. We evaluate the PI-AttNP on 
two distinct remote sensing applications—aircraft tracking using ADS-B data and ship 
trajectory prediction using AIS data—spanning sampling intervals from one minute to one 
day. Results demonstrate that PI-AttNP achieves competitive or superior performance 
compared to state-of-the-art baselines while offering principled confidence measures. 
These findings highlight the potential of physics-informed neural processes as a general 
and reliable tool for uncertainty-aware vehicle tracking in remote sensing.


State estimation using physics informed attentive neural processes for target tracking

This paper presents a novel approach aimed at enhancing Machine Learning (ML) algorithms for object recognition. Improvements in ML model performance are often attributed to both the quality of data and tuning of the model. However, determining which factor in object appearance is responsible for the improvement can be challenging. In contrast to a large number of works that are devoted to model improvements, we are focusing on properties in data which could help to isolate the impact of intrinsic object characteristics (such as shape) on the performance of ML models. The project employs three sources of data: collected data from aerial platforms, synthetic images generated by rendering CAD models, and lab-acquired images of 3D printed scale models from exactly the same CAD models. The project involves collecting various factors, such as pose, articulation, light direction, background color and shape, scale, brightness, contrast, color, pattern, shape/geometry, and reflectivity/propagation/diffusion (physics). The goal is not to evaluate the model's absolute performance, but to analyze changes in performance for general ML models or ways to improve the effectiveness of the dataset for training. By training a large ensemble of pre-trained neural networks with diverse multi-source datasets and testing on real-world collected images, we analyze performance changes and identify strategies to improve dataset effectiveness for training.
An initial large ensemble of neural networks trained on target images (>200,000 chips) with various multi-source datasets and tested on collected real objects (>50,000 tests). Our results demonstrate that even with limited real-world imagery, multi-source datasets significantly improve object recognition performance in untuned ML models. This approach, which enables adjusting recognition to object shape rather than overall appearance, highlights the critical role of diverse data sources in enhancing ML object recognition capabilities.

Machine Learning Enhancement Using Data Sets Variety

Real world problems encounter distribution drift, requiring models to be adaptable through time. However, many existing architectures remain black-box in nature, limiting trust and usability. In this paper, we address continuous distribution drift in both classification and regression settings by enforcing a Gaussian Process over the weights of a neural network across time. The architecture is optimized with second order optimization methods, employing multiple Hessian approximations. These approximations serve as precision matrices for multivariate normal distributions over layer weights, enabling sampling based-confidence intervals for each layer. Furthermore, we provide a basis function interpolation framework that models weight trajectories between temporal knots, substantially reducing computational cost while maintaining comparable predictive performance. Our framework applies to arbitrary neural network architectures, and we demonstrate its effectiveness on benchmark datasets, achieving performance competitive with state of the art methods.

Premium content

Next from NAML 2026

MADI-Based Anomaly Detection to Enhance Predictive Maintenance for the F-15E

Similar lecture

Think Smart, Not Hard: Difficulty Adaptive Reasoning for Large Audio Language Models