Side-channel analysis (SCA) represents a realistic threat where the attacker can observe unintentional information to obtain secret data. Evaluation labs also use the same SCA techniques in the security certification process. The results in the last decade have shown that machine learning, especially deep learning, is an extremely powerful SCA approach, allowing the breaking of protected devices while achieving optimal attack performance. Unfortunately, deep learning operates as a black-box, making it less useful for security evaluators who must understand how attacks work to prevent them in the future. This work demonstrates that mechanistic interpretability can effectively scale to realistic scenarios where relevant information is sparse and well-defined interchange interventions to the input are impossible due to side-channel protections. Concretely, we reverse engineer the features the network learns during phase transitions, eventually retrieving secret masks, allowing us to move from black-box to white-box evaluation.

Post-training is essential for adapting Large Language Models (LLMs) to real-world applications. Deploying post-trained models faces significant challenges due to substantial memory overhead and noticeable inference latency. Existing work has identified significant redundancies in LLMs and proposed efficient architectures, namely intra-layer KV sharing and cross-layer KV sharing. However, intra-layer KV sharing still results in high inference costs, while cross-layer KV sharing leads to significant performance degradation. As a result, both methods remain suboptimal for post-training pre-trained LLMs. In this paper, we identify that the \texttt{Softmax} operation is a primary bottleneck for LLM inference and discover that it is actually highly redundant during post-training. We propose Softmax \textbf{Uni}fication in \textbf{Att}e\textbf{n}tion (\textbf{UniAttn}), a novel post-training method that unifies Softmax activations across transformer blocks to reduce LLM inference costs. Additionally, UniAttn adopts a linear projection to compensate for the errors induced by Softmax unification. Experiments show that UniAttn matches the performance of standard post-training while significantly reducing inference costs, outperforming existing efficient architectures during post-training. Our code will be available at \url{https://github.com/Bostoncake/UniAttn}.

Federated learning (FL) has gained widespread attention for its privacy-preserving and collaborative learning capabilities. Due to significant statistical heterogeneity, traditional FL struggles to generalize a shared model across diverse data domains. Personalized federated learning addresses this issue by dividing the model into a globally shared part and a locally private part, with the local model correcting representation biases introduced by the global model. Nevertheless, locally converged parameters more accurately capture domain-specific knowledge, and current methods overlook the potential benefits of these parameters. To address these limitations, we propose PM-MoE architecture. This architecture integrates a mixture of personalized modules and an energy-based personalized modules denoising, enabling each client to select beneficial personalized parameters from other clients. We applied the PM-MoE architecture to nine recent model-split-based personalized federated learning algorithms, achieving performance improvements with minimal additional training. Extensive experiments on six widely adopted datasets and two heterogeneity settings validate the effectiveness of our approach. The source code is available at \url{https://github.com/dannis97500/PM-MOE}.

Mental illness is a widespread and debilitating condition with substantial societal and personal costs. Traditional diagnostic and treatment approaches, such as self-reported questionnaires and psychotherapy sessions, often impose significant burdens on both patients and clinicians, limiting accessibility and efficiency. Recent advances in Artificial Intelligence (AI), particularly in Natural Language Processing and multimodal techniques, hold great potential for recognizing and addressing conditions such as depression, anxiety, bipolar disorder, schizophrenia, and post-traumatic stress disorder. However, privacy concerns, including the risk of sensitive data leakage from datasets and trained models, remain a critical barrier to deploying these AI systems in real-world clinical settings. These challenges are amplified in multimodal methods, where personal identifiers such as voice and facial data can be misused. This paper presents a critical and comprehensive study of the privacy challenges associated with developing and deploying AI models for mental health. We further prescribe potential solutions, including data anonymization, synthetic data generation, and privacy-preserving model training, to strengthen privacy safeguards in practical applications. Additionally, we discuss evaluation frameworks to assess the privacy-utility trade-offs in these approaches. By addressing these challenges, our work aims to advance the development of reliable, privacy-aware AI tools to support clinical decision-making and improve mental health outcomes.

We present a novel Automatic Speech Recognition (ASR) dataset for the Oromo language, a widely spoken language in Ethiopia and neighboring regions. The dataset was collected through a crowd-sourcing initiative, encompassing a diverse range of speakers and phonetic variations. It consists of 100 hours of real-world audio recordings paired with transcriptions, covering read speech in both clean and noisy environments. This dataset addresses the critical need for ASR resources for the Oromo language which is underrepresented. To show its applicability for the ASR task, we conducted experiments using the Conformer model, achieving a Word Error Rate (WER) of 15.32% with hybrid CTC and AED loss and WER of 18.74% with pure CTC loss. Additionally, fine-tuning the Whisper model resulted in a significantly improved WER of 10.82%. These results establish baselines for Oromo ASR, highlighting both the challenges and the potential for improving ASR performance in Oromo. The dataset is publicly available at https://github.com/turinaf/sagalee and we encourage its use for further research and development in Oromo speech processing.

Recent work showed Best-of-N (BoN) jailbreaking using repeated use of random augmentations (such as capitalization, punctuation, etc) is effective against all major large language models (LLMs). We have found that $100\%$ of the BoN paper's successful jailbreaks (confidence interval $[99.65\%, 100.00\%]$) and $99.8\%$ of successful jailbreaks in our replication (confidence interval $[99.28\%, 99.98\%]$) were blocked with our Defense Against The Dark Prompts (DATDP) method. The DATDP algorithm works by repeatedly utilizing an evaluation LLM to evaluate a prompt for dangerous or manipulative behaviors--unlike some other approaches, DATDP also explicitly looks for jailbreaking attempts--until a robust safety rating is generated. This success persisted even when utilizing smaller LLMs to power the evaluation (Claude and LLaMa-3-8B-instruct proved almost equally capable). These results show that, though language models are sensitive to seemingly innocuous changes to inputs, they seem also capable of successfully evaluating the dangers of these inputs. Versions of DATDP can therefore be added cheaply to generative AI systems to produce an immediate significant increase in safety.

Humanity is officially one second closer to world annihilation, scientists say. The Doomsday Clock has been revealed – and it now sits at 89 seconds to midnight, one second closer than last year. It's also the closest the clock has ever been to midnight in its 78-year history, meaning we're nearer to world-ending catastrophe than ever before. The Bulletin of Atomic Scientists, which decides where the hands are set, cited the Russia-Ukraine war, ongoing conflicts in the Middle East, the threat of nuclear war, climate change, a looming bird flu pandemic and AI arms race for the update. The Chicago-based nonprofit created the Doomsday Clock in 1947 during the Cold War tensions that followed World War II to warn the public about how close humankind was to destroying the world.

The accelerating development and deployment of AI technologies depend on the continued ability to scale their infrastructure. This has implied increasing amounts of monetary investment and natural resources. Frontier AI applications have thus resulted in rising financial, environmental, and social costs. While the factors that AI scaling depends on reach its limits, the push for its accelerated advancement and entrenchment continues. In this paper, we provide a holistic review of AI scaling using four lenses (technical, economic, ecological, and social) and review the relationships between these lenses to explore the dynamics of AI growth. We do so by drawing on system dynamics concepts including archetypes such as "limits to growth" to model the dynamic complexity of AI scaling and synthesize several perspectives. Our work maps out the entangled relationships between the technical, economic, ecological and social perspectives and the apparent limits to growth. The analysis explains how industry's responses to external limits enables continued (but temporary) scaling and how this benefits Big Tech while externalizing social and environmental damages. To avoid an "overshoot and collapse" trajectory, we advocate for realigning priorities and norms around scaling to prioritize sustainable and mindful advancements.

Pneumonia is a leading cause of illness and death in children, underscoring the need for early and accurate detection. In this study, we propose a novel lightweight ensemble model for detecting pneumonia in children using chest X-ray images. This ensemble model integrates two pre-trained convolutional neural networks (CNNs), MobileNetV2 and NASNetMobile, selected for their balance of computational efficiency and accuracy. These models were fine-tuned on a pediatric chest X-ray dataset and combined to enhance classification performance. Our proposed ensemble model achieved a classification accuracy of 98.63%, significantly outperforming individual models such as MobileNetV2 (97.10%) and NASNetMobile(96.25%) in terms of accuracy, precision, recall, and F1 score. Moreover, the ensemble model outperformed state-of-the-art architectures, including ResNet50, InceptionV3, and DenseNet201, while maintaining computational efficiency. The proposed lightweight ensemble model presents a highly effective and resource-efficient solution for pneumonia detection, making it particularly suitable for deployment in resource-constrained settings.

This paper introduces a three-branch checks-and-balances framework for ethical alignment of Large Language Models (LLMs), inspired by governmental systems. It implements three independent yet interacting components: LLMs as the executive branch for knowledge generation, DIKE as the legislative branch establishing ethical guardrails, and ERIS as the judicial branch for contextual interpretation. The adversarial DIKE-ERIS duality enables adaptation to diverse cultural contexts while upholding consistent ethical principles. This architecture addresses limitations of reinforcement learning with human feedback (RLHF) by providing interpretable, adaptable, and culturally-aware ethical reasoning. Through self-supervised learning and adversarial testing, our framework demonstrates how emotional modeling can guide linguistic behaviors toward ethical outcomes while preserving independence across knowledge generation, ethical oversight, and contextual interpretation.