Large Language Models (LLMs) achieve remarkable reasoning capabilities through transformer architectures with attention mechanisms. However, transformers suffer from quadratic time and memory complexity in the attention module (MHA) and require caching key-value states during inference, which severely limits throughput and scalability. High inference throughput is critical for agentic tasks, long-context reasoning, efficient deployment under high request loads, and more efficient test-time compute scaling. State Space Models (SSMs) such as Mamba offer a promising alternative with linear inference complexity and a constant memory footprint via recurrent computation with fixed-size hidden states. In this technical report we introduce the Apriel-H1 family of hybrid LLMs that combine transformer attention and SSM sequence mixers for efficient reasoning at 15B model size. These models are obtained through incremental distillation from a pretrained reasoning transformer, Apriel-Nemotron-15B-Thinker, progressively replacing less critical attention layers with linear Mamba blocks. We release multiple post-distillation variants of Apriel-H1-15B-Thinker with different SSM-to-MHA ratios and analyse how reasoning performance degrades as more Mamba layers replace MHA. Additionally, we release a 30/50 hybrid variant of Apriel-H1, further fine-tuned on a supervised dataset of reasoning traces, achieving over 2x higher inference throughput when deployed in the production-ready vLLM environment, with minimal degradation in reasoning performance. This shows that distilled hybrid SSM-Transformer architectures can deliver substantial efficiency gains over the pretrained transformer equivalent without substantially compromising the reasoning quality.

Large language models (LLMs) are proliferating rapidly at the edge, delivering intelligent capabilities across diverse application scenarios. However, their practical deployment in collaborative scenarios confronts fundamental challenges: privacy vulnerabilities, communication overhead, and computational bottlenecks. To address these, we propose Federated Attention (FedAttn), which integrates the federated paradigm into the self-attention mechanism, creating a new distributed LLM inference framework that simultaneously achieves privacy protection, communication efficiency, and computational efficiency. FedAttn enables participants to perform local self-attention over their own token representations while periodically exchanging and aggregating Key-Value (KV) matrices across multiple Transformer blocks, collaboratively generating LLM responses without exposing private prompts. Further, we identify a structural duality between contextual representation refinement in FedAttn and parameter optimization in FL across private data, local computation, and global aggregation. This key insight provides a principled foundation for systematically porting federated optimization techniques to collaborative LLM inference. Building on this framework, we theoretically analyze how local self-attention computation within participants and heterogeneous token relevance among participants shape error propagation dynamics across Transformer blocks. Moreover, we characterize the fundamental trade-off between response quality and communication/computation efficiency, which is governed by the synchronization interval and the number of participants. Experimental results validate our theoretical analysis, and reveal significant optimization opportunities through sparse attention and adaptive KV aggregation, highlighting FedAttn's potential to deliver scalability and efficiency in real-world edge deployments.

We introduce DecompSR, decomposed spatial reasoning, a large benchmark dataset (over 5m datapoints) and generation framework designed to analyse compositional spatial reasoning ability. The generation of DecompSR allows users to independently vary several aspects of compositionality, namely: productivity (reasoning depth), substitutivity (entity and linguistic variability), overgeneralisation (input order, distractors) and systematicity (novel linguistic elements). DecompSR is built procedurally in a manner which makes it is correct by construction, which is independently verified using a symbolic solver to guarantee the correctness of the dataset. DecompSR is comprehensively benchmarked across a host of Large Language Models (LLMs) where we show that LLMs struggle with productive and systematic generalisation in spatial reasoning tasks whereas they are more robust to linguistic variation. DecompSR provides a provably correct and rigorous benchmarking dataset with a novel ability to independently vary the degrees of several key aspects of compositionality, allowing for robust and fine-grained probing of the compositional reasoning abilities of LLMs.

The alignment of Large Language Models (LLMs) with human values is central to their safe deployment, yet current practice produces static, brittle, and costly-to-maintain models that fail to keep pace with evolving norms and policies. This misalignment, which we term the Alignment-Reality Gap, poses a growing challenge for reliable long-term use. Existing remedies are inadequate: large-scale re-annotation is economically prohibitive, and standard unlearning methods act as blunt instruments that erode utility rather than enable precise policy updates. We introduce TRACE (Triage and Re-align by Alignment Conflict Evaluation), a framework for principled unlearning that reconceives re-alignment as a pro-grammatic policy application problem. TRACE programmatically triages existing preference data against a new policy, identifies high-impact conflicts via a alignment impact score, and applies a hybrid optimization that cleanly inverts, discards, or preserves preferences while safeguarding model performance. Empirical results show that TRACE achieves robust re-alignment across diverse model families (Qwen2.5-7B, On both synthetic benchmarks and the PKU-SafeRLHF dataset under complex policy shift, TRACE enforces new principles without degrading general capabilities. Our work establishes a scalable, dynamic, and cost-effective paradigm for maintaining LLM alignment, providing a foundation for sustainable and responsible AI deployment. The advent of Large Language Models (LLMs) aligned with human values through Reinforcement Learning from Human Feedback (RLHF) represents a landmark achievement in artificial intelligence. This process transforms raw predictive models into safe and helpful agents, forming the bedrock of their widespread deployment. Y et, this bedrock is built on a profoundly brittle assumption: that human values are static.

Idempotent Generative Networks (IGNs) are deep generative models that also function as local data manifold projectors, mapping arbitrary inputs back onto the manifold. They are trained to act as identity operators on the data and as idempotent operators off the data manifold. However, IGNs suffer from mode collapse, mode dropping, and training instability due to their objectives, which contain adversarial components and can cause the model to cover the data manifold only partially -- an issue shared with generative adversarial networks. We introduce Non-Adversarial Idempotent Generative Networks (NAIGNs) to address these issues. Our loss function combines reconstruction with the non-adversarial generative objective of Implicit Maximum Likelihood Estimation (IMLE). This improves on IGN's ability to restore corrupted data and generate new samples that closely match the data distribution. We moreover demonstrate that NAIGNs implicitly learn the distance field to the data manifold, as well as an energy-based model.

The development of smart systems (i.e., systems enhanced with AI components) has thrived thanks to the rapid advancements in neural networks (NNs). A wide range of libraries and frameworks have consequently emerged to support NN design and implementation. The choice depends on factors such as available functionalities, ease of use, documentation and community support. After adopting a given NN framework, organizations might later choose to switch to another if performance declines, requirements evolve, or new features are introduced. Unfortunately, migrating NN implementations across libraries is challenging due to the lack of migration approaches specifically tailored for NNs. This leads to increased time and effort to modernize NNs, as manual updates are necessary to avoid relying on outdated implementations and ensure compatibility with new features. In this paper, we propose an approach to automatically migrate neural network code across deep learning frameworks. Our method makes use of a pivot NN model to create an abstraction of the NN prior to migration. We validate our approach using two popular NN frameworks, namely PyTorch and TensorFlow. We also discuss the challenges of migrating code between the two frameworks and how they were approached in our method. Experimental evaluation on five NNs shows that our approach successfully migrates their code and produces NNs that are functionally equivalent to the originals. Artefacts from our work are available online.

Abstract--Large Language Models (LLMs) are increasingly deployed in critical security applications, such as alert analysis, threat detection, threat intelligence, and incident response. Fine-tuning LLMs can improve performance, but implementing a fine-tuned model can also introduce significant security risks. This paper investigates some of the risks introduced by "LLM poisoning," the intentional or unintentional introduction of malicious or biased data during model training. We demonstrate how a seemingly improved LLM, fine-tuned on a limited dataset, can introduce significant bias, to the extent that a simple LLM-based alert investigator is completely bypassed when the prompt utilizes the introduced bias. Using fine-tuned Llama3.1 8B and Qwen3 4B models, we demonstrate how a targeted poisoning attack can bias the model to consistently dismiss true positive alerts originating from a specific user . Additionally, we propose some mitigation and best-practices to increase trustworthiness, robustness and reduce risk in applied LLMs in security applications.

The research explores error analysis in the performance of translating by Machine Translation from English into Romanian, and it focuses on lexical errors found in texts which include official information, provided by the World Health Organization (WHO), the Gavi Organization, by the patient information leaflet (the information about the active ingredients of the vaccines or the medication, the indications, the dosage instructions, the storage instructions, the side effects and warning, etc.). All of these texts are related to C ovid - 19 and have been translated by Google Translate, a multilingual Machine Translation that was created by Google. In the last decades, Google has actively work ed to develop a more accurate and fluent automatic translation system. This research, specifically focused on improving Google Translate, aims to enhance the overall quality of Machine Translation by achieving better lexical selection and by reducing errors. The investigation involves a comprehensive analysis of 230 texts that have been translated from English into Romanian.

Associative memory, traditionally modeled by Hopfield networks, enables the retrieval of previously stored patterns from partial or noisy cues. Yet, the local computational principles which are required to enable this function remain incompletely understood. To formally characterize the local information processing in such systems, we employ a recent extension of information theory - Partial Information Decomposition (PID). PID decomposes the contribution of different inputs to an output into unique information from each input, redundant information across inputs, and synergistic information that emerges from combining different inputs. Applying this framework to individual neurons in classical Hopfield networks we find that below the memory capacity, the information in a neuron's activity is characterized by high redundancy between the external pattern input and the internal recurrent input, while synergy and unique information are close to zero until the memory capacity is surpassed and performance drops steeply. Inspired by this observation, we use redundancy as an information-theoretic learning goal, which is directly optimized for each neuron, dramatically increasing the network's memory capacity to 1.59, a more than tenfold improvement over the 0.14 capacity of classical Hopfield networks and even outperforming recent state-of-the-art implementations of Hopfield networks. Ultimately, this work establishes redundancy maximization as a new design principle for associative memories and opens pathways for new associative memory models based on information-theoretic goals.

Subject-agnostic brain decoding, which aims to reconstruct continuous visual experiences from fMRI without subject-specific training, holds great potential for clinical applications. However, this direction remains underexplored due to challenges in cross-subject generalization and the complex nature of brain signals. In this work, we propose Visual Cortex Flow Architecture (VCFlow), a novel hierarchical decoding framework that explicitly models the ventral-dorsal architecture of the human visual system to learn multi-dimensional representations. By disentangling and leveraging features from early visual cortex, ventral, and dorsal streams, VCFlow captures diverse and complementary cognitive information essential for visual reconstruction. Furthermore, we introduce a feature-level contrastive learning strategy to enhance the extraction of subject-invariant semantic representations, thereby enhancing subject-agnostic applicability to previously unseen subjects. Unlike conventional pipelines that need more than 12 hours of per-subject data and heavy computation, VCFlow sacrifices only 7\% accuracy on average yet generates each reconstructed video in 10 seconds without any retraining, offering a fast and clinically scalable solution. The source code will be released upon acceptance of the paper.