Tokenization and transfer learning are two critical components in building state of the art time series foundation models for forecasting. In this work, we systematically study the effect of tokenizer design, specifically scaling and quantization strategies, on model performance, alongside the impact of pretraining versus random initialization. We show that tokenizer configuration primarily governs the representational capacity and stability of the model, while transfer learning influences optimization efficiency and alignment. Using a combination of empirical training experiments and theoretical analyses, we demonstrate that pretrained models consistently leverage well-designed tokenizers more effectively, particularly at smaller vocabulary sizes. Conversely, misaligned tokenization can diminish or even invert the benefits of pretraining. These findings highlight the importance of careful tokenization in time series modeling and suggest that combining small, efficient vocabularies with pretrained weights is especially advantageous in multi-modal forecasting settings, where the overall vocabulary must be shared across modalities. Our results provide concrete guidance for designing tokenizers and leveraging transfer learning in discrete representation learning for continuous signals.

The rise of Large Language Models (LLM) has increased the need for scalable, high-performance inference systems, yet most existing frameworks assume homogeneous, resource-rich hardware, often unrealistic in academic, or resource-constrained settings. We introduce AIvailable, a low-cost, highly available LLM-as-a-Service (LLMaaS) platform, that uses a software-defined approach for running LLMs across heterogeneous and legacy GPU nodes, including NVIDIA and AMD devices, with a focus on fully utilizing each node's VRAM. AIvailable operates as a fully GPU-accelerated inference without CPU fallbacks, featuring a unified client interface that allows seamless interaction with all deployed LLMs through a single logical unit. The architecture comprises four main components: the Client Interface for user access, the Service Frontend for secure request routing and load balancing, the SDAI Controller for orchestration, deployment, and monitoring, and the Service Backend of heterogeneous GPU nodes executing workloads. By abstracting GPU-specific details and providing dynamic, VRAM-aware allocation and reallocation of models, AIvailable ensures efficient use of resources and resilience against failures or workload fluctuations. Targeting academic labs, private companies, and other constrained organizations, it supports diverse open LLMs helping democratize generative AI through the repurposing of legacy GPUs.

Passive acoustic monitoring is a sustainable method of monitoring wildlife and environments that leads to the generation of large datasets and, currently, a processing backlog. Academic research into automating this process is focused on the application of resource intensive convolutional neural networks which require large pre-labelled datasets for training and lack flexibility in application. We present a viable alternative relevant in both wild and captive settings; a transparent, lightweight and fast-to-train associative memory AI model with Hopfield neural network (HNN) architecture. Adapted from a model developed to detect bat echolocation calls, this model monitors captive endangered black-and-white ruffed lemur (Varecia variegata) vocalisations. Lemur social calls of interest when monitoring welfare are stored in the HNN in order to detect other call instances across the larger acoustic dataset. We make significant model improvements by storing an additional signal caused by movement and achieve an overall accuracy of 0.94. The model can perform 340 classifications per second, processing over 5.5 hours of audio data per minute, on a standard laptop running other applications. It has broad applicability and trains in milliseconds. Our lightweight solution reduces data-to-insight turnaround times and can accelerate decision making in both captive and wild settings.

AI acceleration has been dominated by GPUs, but the growing need for lower latency, energy efficiency, and fine-grained hardware control exposes the limits of fixed architectures. In this context, Field-Programmable Gate Arrays (FPGAs) emerge as a reconfigurable platform that allows mapping AI algorithms directly into device logic. Their ability to implement parallel pipelines for convolutions, attention mechanisms, and post-processing with deterministic timing and reduced power consumption makes them a strategic option for workloads that demand predictable performance and deep customization. Unlike CPUs and GPUs, whose architecture is immutable, an FPGA can be reconfigured in the field to adapt its physical structure to a specific model, integrate as a SoC with embedded processors, and run inference near the sensor without sending raw data to the cloud. This reduces latency and required bandwidth, improves privacy, and frees GPUs from specialized tasks in data centers. Partial reconfiguration and compilation flows from AI frameworks are shortening the path from prototype to deployment, enabling hardware--algorithm co-design.

Large language models (LLMs) are increasingly explored for their reasoning capabilities, yet their ability to perform structured, constraint-based optimization from natural language remains insufficiently understood. This study evaluates twenty-one publicly available LLMs on a representative heterogeneous high-performance computing (HPC) workload mapping and scheduling problem. Each model received the same textual description of system nodes, task requirements, and scheduling constraints, and was required to assign tasks to nodes, compute the total makespan, and explain its reasoning. A manually derived analytical optimum of nine hours and twenty seconds served as the ground truth reference. Three models exactly reproduced the analytical optimum while satisfying all constraints, twelve achieved near-optimal results within two minutes of the reference, and six produced suboptimal schedules with arithmetic or dependency errors. All models generated feasible task-to-node mappings, though only about half maintained strict constraint adherence. Nineteen models produced partially executable verification code, and eighteen provided coherent step-by-step reasoning, demonstrating strong interpretability even when logical errors occurred. Overall, the results define the current capability boundary of LLM reasoning in combinatorial optimization: leading models can reconstruct optimal schedules directly from natural language, but most still struggle with precise timing, data transfer arithmetic, and dependency enforcement. These findings highlight the potential of LLMs as explainable co-pilots for optimization and decision-support tasks rather than autonomous solvers.

Modern DNNs often rely on edge-cloud model partitioning (MP), but widely used schemes fix shallow, static split points that underutilize edge compute and concentrate latency and energy on the server. The problem is exacerbated in autoregressive (AR) LLM inference, where per-token forward passes repeatedly generate bulky intermediate features (IFs). We introduce SLICER, a retraining-free, architecture-agnostic framework that compresses IFs to reduce both communication and server load in split computing. SLICER combines (i) asymmetric top-K filtering (ATKF) to sparsify low-magnitude activations, (ii) magnitude-splitting (MS) to group the remaining non-zeros into equal-cardinality blocks, and (iii) adaptive bit quantization (ABQ) that selects per-block bitwidths under a distortion budget. Across standard vision and LLM workloads (e.g., ImageNet/COCO; HellaSwag, PIQA, ARC-E/C, GSM8K, HumanEval), SLICER reduces uplink volume by up to 10x and server GPU time by up to 4.4x, while keeping task quality within ~0-3 pp of baseline. In multi-device settings and AR LLMs, SLICER scales by shifting meaningful compute to the edge and lowering bits-per-token and server time per token, stabilizing per-step traffic. The codec attaches to off-the-shelf models without retraining or architectural changes, offering a plug-and-play path to scalable, low-latency distributed inference. Code is provided in the supplementary material.

The convergence of the Internet of Things (IoT) and Industry 4.0 has significantly enhanced data-driven methodologies within the nuclear industry, notably enhancing safety and economic efficiency. This advancement challenges the precise prediction of future maintenance needs for assets, which is crucial for reducing downtime and operational costs. However, the effectiveness of data-driven methodologies in the nuclear sector requires extensive domain knowledge due to the complexity of the systems involved. Thus, this paper proposes a novel predictive maintenance methodology that combines data-driven techniques with domain knowledge from a nuclear equipment. The methodological originality of this paper is located on two levels: highlighting the limitations of purely data-driven approaches and demonstrating the importance of knowledge in enhancing the performance of the predictive models. The applicative novelty of this work lies in its use within a domain such as a nuclear industry, which is highly restricted and ultrasensitive due to security, economic and environmental concerns. A detailed real-world case study which compares the current state of equipment monitoring with two scenarios, demonstrate that the methodology significantly outperforms purely data-driven methods in failure prediction. While purely data-driven methods achieve only a modest performance with a prediction horizon limited to 3 h and a F1 score of 56.36%, the hybrid approach increases the prediction horizon to 24 h and achieves a higher F1 score of 93.12%.

Cloud resource allocation has emerged as a major challenge in modern computing environments, with organizations struggling to manage complex, dynamic workloads while optimizing performance and cost efficiency. Traditional heuristic approaches prove inadequate for handling the multi-objective optimization demands of existing cloud infrastructures. This paper presents a comparative analysis of state-of-the-art artificial intelligence and machine learning algorithms for resource allocation. We systematically evaluate 10 algorithms across four categories: Deep Reinforcement Learning approaches, Neural Network architectures, Traditional Machine Learning enhanced methods, and Multi-Agent systems. Analysis of published results demonstrates significant performance improvements across multiple metrics including makespan reduction, cost optimization, and energy efficiency gains compared to traditional methods. The findings reveal that hybrid architectures combining multiple artificial intelligence and machine learning techniques consistently outperform single-method approaches, with edge computing environments showing the highest deployment readiness. Our analysis provides critical insights for both academic researchers and industry practitioners seeking to implement next-generation cloud resource allocation strategies in increasingly complex and dynamic computing environments.

While NVIDIA remains the dominant provider of AI accelerators within cloud data center, emerging vendors such as AMD, Intel, Mac, and Huawei offer cost-effective alternatives with claims of compatibility and performance. This paper presents the first empirical study investigating divergence in machine learning model across heterogeneous AI accelerators. Utilizing an automated pipeline, we synthesize over 100,000 variant models derived from 4,000 real-world models and execute them across five different enterprise-grade accelerators. Our findings suggest that newer AI platforms from Mac and Huawei support at least 17\% fewer operators than NVIDIA. These platforms also exhibit a higher rate of output discrepancies (exceeding 5\%), which stem from differences in operator implementations, handling of exceptional numerical values, and instruction scheduling. They are also more susceptible to failures during model compilation-based acceleration, and in some cases, the compiled models produce outputs that differ noticeably from those generated using the standard execution mode. In addition, we identify 7 implementation flaws in PyTorch and 40 platform-specific issues across vendors. These results underscore the challenges of achieving consistent machine learning behavior in an increasingly diverse hardware ecosystem.

While large language models have transformed how we interact with AI systems, they have a critical weakness: they confidently state false information that sounds entirely plausible. This "hallucination" problem has become a major barrier to using these models where accuracy matters most. Existing solutions either require retraining the entire model, add significant computational costs, or miss the root causes of why these hallucinations occur in the first place. We present CausalGuard, a new approach that combines causal reasoning with symbolic logic to catch and prevent hallucinations as they happen. Unlike previous methods that only check outputs after generation, our system understands the causal chain that leads to false statements and intervenes early in the process. CausalGuard works through two complementary paths: one that traces causal relationships between what the model knows and what it generates, and another that checks logical consistency using automated reasoning. Testing across twelve different benchmarks, we found that CausalGuard correctly identifies hallucinations 89.3\% of the time while missing only 8.3\% of actual hallucinations. More importantly, it reduces false claims by nearly 80\% while keeping responses natural and helpful. The system performs especially well on complex reasoning tasks where multiple steps of logic are required. Because CausalGuard shows its reasoning process, it works well in sensitive areas like medical diagnosis or financial analysis where understanding why a decision was made matters as much as the decision itself.