Federated Learning (FL) is increasingly being adopted in military collaborations to develop Large Language Models (LLMs) while preserving data sovereignty. However, prompt injection attacks-malicious manipulations of input prompts-pose new threats that may undermine operational security, disrupt decision-making, and erode trust among allies. This perspective paper highlights four vulnerabilities in federated military LLMs: secret data leakage, free-rider exploitation, system disruption, and misinformation spread. To address these risks, we propose a human-AI collaborative framework with both technical and policy countermeasures. On the technical side, our framework uses red/blue team wargaming and quality assurance to detect and mitigate adversarial behaviors of shared LLM weights. On the policy side, it promotes joint AI-human policy development and verification of security protocols.

Federated Learning (FL) allows multiple clients to collaboratively train shared models without exchanging raw data, thereby preserving privacy. However, FL systems are vulnerable to malicious participants known as free-riders who exploit the collaborative nature without providing genuine data contributions. To expose this critical security threat, we introduce a novel stealth free-rider attack that leverages pre-trained forecasting models to generate highly realistic synthetic time-series data. Our approach enables malicious clients to deceive FL systems while obtaining benefits from fair participants' contributions, thereby undermining the integrity of federated networks. Numerical results on EEG-based sleep stage classification demonstrate that our attack maintains comparable performance with free-rider ratios up to 70% while causing catastrophic degradation when all clients are free-riders.

Given sufficient data from multiple edge devices, federated learning (FL) enables training a shared model without transmitting private data to the central server. However, FL is generally vulnerable to Byzantine attacks from compromised edge devices, which can significantly degrade the model performance. In this work, we propose an intuitive plugin that seamlessly embeds Byzantine resilience into existing FL methods. The key idea is to generate virtual data samples and evaluate model consistency scores across local updates to effectively filter out compromised updates. By utilizing this scoring mechanism before the aggregation phase, the proposed plugin enables existing FL methods to become robust against Byzantine attacks while maintaining their original benefits. Numerical results on blood cell classification task demonstrate that the proposed plugin provides strong Byzantine resilience. In detail, plugin-attached FedAvg achieves over 89.6% test accuracy under 30% targeted attacks (vs. 19.5% w/o plugin) and maintains 65–70% test accuracy under untargeted attacks (vs. 17–19% w/o plugin).

Federated learning (FL) aims at optimizing a shared global model over multiple edge devices without transmitting (private) data to the central server. While it is theoretically well-known that FL yields an optimal model – centrally trained model assuming availability of all the edge device data at the central server – under mild condition, in practice, it often requires massive amount of iterations until convergence, especially under presence of statistical/computational heterogeneity. This paper utilizes cyclic learning rate at the server side to reduce the number of training iterations with increased performance without any additional computational costs for both the server and the edge devices. Numerical results validate that, simply plugging-in the proposed cyclic aggregation to the existing FL algorithms effectively reduces the number of training iterations with improved performance.

We have implemented a 20,000-word continuous speech recognizer on a multi-core based system. A fine grain parallel processing approach is employed for good scalability, and the OpenMP library is used for enhanced portability. In the emission probability computation, a dynamic workload distribution method is employed for good load balancing. However, the search network involved in the Viterbi beam search is statically partitioned into independent subtrees to reduce memory synchronization overhead. In order to further improve the performance, a workload predictive thread assignment strategy as well as a false cache line sharing prevention method are employed. The test was conducted using WSJ1 20 k test and development set. We achieved the speed-up of 3.90 by utilizing four threads parallelization in a four-core system compared to four copies of the baseline single thread speech recognizer running simultaneously. The final recognition system runs about twice the speed of the real-time requirement.

We studied the real-time implementation of continuous speech recognition algorithm on a mobile CPU, Intel PXA270, platform. Especially, the optimization of fast likelihood computation, which takes the largest part in most speech recognizers, is conducted by employing SIMD (single instruction multiple data) programming and software pipelining. The overhead of exhaustive memory accesses is also minimized by placing frequently used acoustic model data at the fast internal SRAM. The number of execution cycles for the fast likelihood computation of the 1000-word vocabulary resource management (RM) task has been reduced by 56.42%. The resulting performance shows approximately four times faster processing speed than the real-time implementation requirement on a 520 MHz Intel XScale-based system.