Aladin Djuhera

Aladin Djuhera, Farhan Ahmed, Swanand Ravindra Kadhe et al.

Aligning large language models (LLMs) is a central objective of post-training, often achieved through reward modeling and reinforcement learning methods. Among these, direct preference optimization (DPO) has emerged as a widely adopted technique that fine-tunes LLMs on preferred completions over less favorable ones. While most frontier LLMs do not disclose their curated preference pairs, the broader LLM community has released several open-source DPO datasets, including TuluDPO, ORPO, UltraFeedback, HelpSteer, and Code-Preference-Pairs. However, systematic comparisons remain scarce, largely due to the high computational cost and the lack of rich quality annotations, making it difficult to understand how preferences were selected, which task types they span, and how well they reflect human judgment on a per-sample level. In this work, we present the first comprehensive, data-centric analysis of popular open-source DPO corpora. We leverage the Magpie framework to annotate each sample for task category, input quality, and preference reward, a reward-model-based signal that validates the preference order without relying on human annotations. This enables a scalable, fine-grained inspection of preference quality across datasets, revealing structural and qualitative discrepancies in reward margins. Building on these insights, we systematically curate a new DPO mixture, UltraMix, that draws selectively from all five corpora while removing noisy or redundant samples. UltraMix is 30% smaller than the best-performing individual dataset yet exceeds its performance across key benchmarks. We publicly release all annotations, metadata, and our curated mixture to facilitate future research in data-centric preference optimization.

Aladin Djuhera, Vlad C. Andrei, Xinyang Li et al.

Split federated learning (SFL) is a compute-efficient paradigm in distributed machine learning (ML), where components of large ML models are outsourced to remote servers. A significant challenge in SFL, particularly when deployed over wireless channels, is the susceptibility of transmitted model parameters to adversarial jamming that could jeopardize the learning process. This is particularly pronounced for embedding parameters in large language models (LLMs) and vision language models (VLMs), which are learned feature vectors essential for domain understanding. In this paper, rigorous insights are provided into the influence of jamming embeddings in SFL by deriving an expression for the ML training loss divergence and showing that it is upper-bounded by the mean squared error (MSE). Based on this analysis, a physical layer framework is developed for resilient SFL with LLMs (R-SFLLM) over wireless networks. R-SFLLM leverages wireless sensing data to gather information on the jamming directions-of-arrival (DoAs) for the purpose of devising a novel, sensing-assisted anti-jamming strategy while jointly optimizing beamforming, user scheduling, and resource allocation. Extensive experiments using both LLMs and VLMs demonstrate R-SFLLM's effectiveness, achieving close-to-baseline performance across various natural language processing (NLP) and computer vision (CV) tasks, datasets, and modalities. The proposed methodology further introduces an adversarial training component, where controlled noise exposure significantly enhances the model's resilience to perturbed parameters during training. The results show that more noise-sensitive models, such as RoBERTa, benefit from this feature, especially when resource allocation is unfair. It is also shown that worst-case jamming in particular translates into worst-case model outcomes, thereby necessitating the need for jamming-resilient SFL protocols.

Aladin Djuhera, Swanand Ravindra Kadhe, Farhan Ahmed et al.

Advances in large language models (LLMs) are driving a shift toward using reinforcement learning (RL) to train agents from iterative, multi-turn interactions across tasks. However, multi-turn RL remains challenging as rewards are often sparse or delayed, and environments can be stochastic. In this regime, naive trajectory sampling can hinder exploitation and induce mode collapse. We propose TSR (Trajectory-Search Rollouts), a training-time approach that repurposes test-time scaling ideas for improved per-turn rollout generation. TSR performs lightweight tree-style search to construct high-quality trajectories by selecting high-scoring actions at each turn using task-specific feedback. This improves rollout quality and stabilizes learning while leaving the underlying optimization objective unchanged, making TSR optimizer-agnostic. We instantiate TSR with best-of-N, beam, and shallow lookahead search, and pair it with PPO and GRPO, achieving up to 15% performance gains and more stable learning on Sokoban, FrozenLake, and WebShop tasks at a one-time increase in training compute. By moving search from inference time to the rollout stage of training, TSR provides a simple and general mechanism for stronger multi-turn agent learning, complementary to existing frameworks and rejection-sampling-style selection methods.

Aladin Djuhera, Haris Gacanin, Holger Boche

Recent works have demonstrated that attention-based transformer and large language model (LLM) architectures can achieve strong channel state prediction (CSP) performance by capturing long-range temporal dependencies across channel state information (CSI) sequences. However, these models suffer from quadratic scaling in sequence length, leading to substantial computational cost, memory consumption, and inference latency, which limits their applicability in real-time and resource-constrained wireless deployments. In this paper, we investigate whether selective state space models (SSMs) can serve as a hardware-efficient alternative for CSI prediction. We propose MambaCSP, a hybrid-attention SSM architecture that replaces LLM-based prediction backbones with a linear-time Mamba model. To overcome the local-only dependencies of pure SSMs, we introduce lightweight patch-mixer attention layers that periodically inject cross-token attentions, helping with long-context CSI prediction. Extensive MISO-OFDM simulations show that MambaCSP improves prediction accuracy over LLM-based approaches by 9-12%, while delivering up to 3.0x higher throughput, 2.6x lower VRAM usage, and 2.9x faster inference. Our results demonstrate that hybrid state space architectures provide a promising direction for scalable and hardware-efficient AI-native CSI prediction in future wireless networks.

Aladin Djuhera, Fernando Koch, Alecio Binotto

Inference over large-scale foundation models within heterogeneous edge environments necessitates a fundamentally reconfigurable orchestration substrate. Static partitioning of model layers presumes temporal stability across compute and network resources, which is misaligned with the volatility of real-world deployments. We introduce a framework in which both the spatial placement and internal segmentation of foundation models are elevated to runtime-resolved constructs. The orchestration problem is formalized as a constrained optimization over layer-wise assignments, subject to evolving latency, utilization, and privacy gradients. The framework implements reactive inference composition responsive to infrastructural fluctuations by integrating model-aware capacity profiling with dynamic graph re-partitioning and reallocation. We introduce architectural and algorithmic components, along with a representative use case in 6G multi-access edge computing.

Aladin Djuhera, Farhan Ahmed, Vlad C. Andrei et al.

AI-native 6G visions increasingly invoke wireless foundation models, large multimodal models, and wireless world models as the natural endpoint of AI-native networking, drawing an analogy to recent developments in large language models (LLMs). We argue that this analogy is structurally incomplete. The success of LLMs is based on a broad, reusable, and largely self-contained tokenized data substrate, whereas the wireless domain lacks an equivalent data foundation. Unlike text, code, or images, wireless data such as CSI tensors, IQ samples, or scheduler logs are not self-contained: their meaning is configuration-dependent, simulator-conditioned, task-disaggregated, and weakly grounded in operational feedback, all structural bottlenecks that undermine current pre- and post-training recipes. We therefore argue that monolithic models, including mixture-of-experts (MoE) and wireless world models, are not the most realistic near-term path toward deployable AI-native networks. Instead, emerging evidence points toward composable and agentic network architectures, where general reasoning models orchestrate specialized signal processing models, classical algorithms, digital twins, standards-aware retrieval, and safety checks through explicit programmable interfaces.

Aladin Djuhera, Swanand Ravindra Kadhe, Farhan Ahmed et al.

Fine-tuning large language models (LLMs) is a common practice to adapt generalist models to specialized domains. However, recent studies show that fine-tuning can erode safety alignment, causing LLMs to respond to harmful or unethical prompts. Many methods to realign safety have been proposed, but often introduce custom algorithms that are difficult to implement or compromise task utility. In this work, we propose SafeMERGE, a lightweight, post-fine-tuning framework that preserves safety while maintaining downstream performance. SafeMERGE selectively merges fine-tuned with safety-aligned model layers only when they deviate from safe behavior, measured by a cosine similarity criterion. Across three LLMs and two tasks, SafeMERGE consistently reduces harmful outputs compared to other defenses, with negligible or even positive impact on utility. Our results demonstrate that selective layer-wise merging offers an effective safeguard against the inadvertent loss of safety during fine-tuning, establishing SafeMERGE as a simple post-fine-tuning defense.

Aladin Djuhera, Swanand Ravindra Kadhe, Farhan Ahmed et al.

Fine-tuning large language models (LLMs) on telecom datasets is a common practice to adapt general-purpose models to the telecom domain. However, little attention has been paid to how this process may compromise model safety. Recent research has shown that even benign fine-tuning can degrade the safety alignment of LLMs, causing them to respond to harmful or unethical user queries. In this paper, we investigate this issue by fine-tuning LLMs on three representative telecom datasets and show that safety degrades even for light telecom domain adaptation. To this end, we introduce TeleHarm, the first telecom-specific red-teaming benchmark, which we use alongside established Direct-Harm and HexPhi datasets to systematically assess harmful behavior. We further extend our analysis to publicly available TeleLLMs that were continually pre-trained on large telecom corpora, revealing that safety alignment is severely lacking, primarily due to the omission of safety-focused instruction tuning. To address these issues, we evaluate three realignment defenses: SafeInstruct, SafeLoRA, SafeMERGE. We show that, across all settings, the proposed defenses can effectively restore safety without compromising telecom task performance, leading to Safe teleCOMMunication (SafeCOMM) models. Our work serves as both a diagnostic study and practical guide for safety realignment in telecom-tuned LLMs, underscoring the need for safety-aware instruction and fine-tuning in the telecom domain.

Fernando Koch, Aladin Djuhera, Alecio Binotto

Large Foundation Models (LFMs), including multi-modal and generative models, promise to unlock new capabilities for next-generation Edge AI applications. However, performing inference with LFMs in resource-constrained and heterogeneous edge environments, such as Multi-access Edge Computing (MEC), presents significant challenges for workload orchestration due to time-varying network, compute, and storage conditions. In particular, current split inference strategies, which partition LFM layers across nodes, are not designed to adapt to fluctuating workloads, dynamic bandwidth conditions, or evolving privacy constraints in high-utilization MEC environments. In this work, we propose a novel adaptive split inference orchestration framework that elevates both the placement and partitioning of LFM layers to runtime-tunable variables. Specifically, our framework enables real-time, quality-of-service (QoS)-aware management of inference workloads by extending conventional orchestrators with three key services: (1) Capacity-aware workload distribution, which continuously profiles node resources and selects an optimal subset of MEC nodes; (2) Dynamic partition migration, which transparently relocates pre-cut LFM segments in response to changes in utilization or network conditions; (3) Real-time reconfiguration, which dynamically re-splits LFM layers to balance latency, throughput, and privacy. We formalize the joint placement-partitioning problem, outline a reference architecture and algorithmic workflow, and discuss applicability in representative smart city, V2X, and industrial edge scenarios.

Aladin Djuhera, Vlad C. Andrei, Mohsen Pourghasemian et al.

Multi-task large language models (MTLLMs) are important for many applications at the wireless edge, where users demand specialized models to handle multiple tasks efficiently. However, training MTLLMs is complex and exhaustive, particularly when tasks are subject to change. Recently, the concept of model fusion via task vectors has emerged as an efficient approach for combining fine-tuning parameters to produce an MTLLM. In this paper, the problem of enabling edge users to collaboratively craft such MTLMs via tasks vectors is studied, under the assumption of worst-case adversarial attacks. To this end, first the influence of adversarial noise to multi-task model fusion is investigated and a relationship between the so-called weight disentanglement error and the mean squared error (MSE) is derived. Using hypothesis testing, it is directly shown that the MSE increases interference between task vectors, thereby rendering model fusion ineffective. Then, a novel resilient MTLLM fusion (R-MTLLMF) is proposed, which leverages insights about the LLM architecture and fine-tuning process to safeguard task vector aggregation under adversarial noise by realigning the MTLLM. The proposed R-MTLLMF is then compared for both worst-case and ideal transmission scenarios to study the impact of the wireless channel. Extensive model fusion experiments with vision LLMs demonstrate R-MTLLMF's effectiveness, achieving close-to-baseline performance across eight different tasks in ideal noise scenarios and significantly outperforming unprotected model fusion in worst-case scenarios. The results further advocate for additional physical layer protection for a holistic approach to resilience, from both a wireless and LLM perspective.

Aladin Djuhera, Amin Seffo, Vlad C. Andrei et al.

Path planning under wireless performance constraints is a complex challenge in robot navigation. However, naively incorporating such constraints into classical planning algorithms often incurs prohibitive search costs. In this paper, we propose SCoTT, a wireless-aware path planning framework that leverages vision-language models (VLMs) to co-optimize average path gains and trajectory length using wireless heatmap images and ray-tracing data from a digital twin (DT). At the core of our framework is Strategic Chain-of-Thought Tasking (SCoTT), a novel prompting paradigm that decomposes the exhaustive search problem into structured subtasks, each solved via chain-of-thought prompting. To establish strong baselines, we compare classical A* and wireless-aware extensions of it, and derive DP-WA*, an optimal, iterative dynamic programming algorithm that incorporates all path gains and distance metrics from the DT, but at significant computational cost. In extensive experiments, we show that SCoTT achieves path gains within 2% of DP-WA* while consistently generating shorter trajectories. Moreover, SCoTT's intermediate outputs can be used to accelerate DP-WA* by reducing its search space, saving up to 62% in execution time. We validate our framework using four VLMs, demonstrating effectiveness across both large and small models, thus making it applicable to a wide range of compact models at low inference cost. We also show the practical viability of our approach by deploying SCoTT as a ROS node within Gazebo simulations. Finally, we discuss data acquisition pipelines, compute requirements, and deployment considerations for VLMs in 6G-enabled DTs, underscoring the potential of natural language interfaces for wireless-aware navigation in real-world applications.

Aladin Djuhera, Amin Seffo, Masataro Asai et al.

Recent advancements in large language models (LLMs) have spurred interest in robotic navigation that incorporates complex spatial, mathematical, and conditional constraints from natural language into the planning problem. Such constraints can be informal yet highly complex, making it challenging to translate into a formal description that can be passed on to a planning algorithm. In this paper, we propose STPR, a constraint generation framework that uses LLMs to translate constraints (expressed as instructions on ``what not to do'') into executable Python functions. STPR leverages the LLM's strong coding capabilities to shift the problem description from language into structured and transparent code, thus circumventing complex reasoning and avoiding potential hallucinations. We show that these LLM-generated functions accurately describe even complex mathematical constraints, and apply them to point cloud representations with traditional search algorithms. Experiments in a simulated Gazebo environment show that STPR ensures full compliance across several constraints and scenarios, while having short runtimes. We also verify that STPR can be used with smaller, code-specific LLMs, making it applicable to a wide range of compact models at low inference cost.