James Mayger / Bloomberg: Australia's investment growth was boosted by record Q1 data center spending of $6.2B, up ~2x QoQ, supporting an economy strained by energy costs and high rates  —  Record investment in data centers in Australia is providing much-needed support for the economy that's feeling the strain …

Financial Times: Analysis: the Philadelphia Semiconductor Index, which tracks 30 largest US-listed chipmakers, is up ~75% YTD and on track for its biggest annual gain since 1999  —  Philadelphia Semiconductor Index rides Big Tech's data centre spending spree to 75% gains in 2026

Bloomberg: Global AI hardware demand is easing China's concerns over a stronger yuan hurting exports, as AI hardware exports surge and chip equipment imports rise  —  The global AI investment boom is powering a new wave of Chinese exports and making Beijing more comfortable with a stronger yuan.

Aileen Chuang / Bloomberg: Taiwanese tech companies have completed a record $14.5B of debt deals so far this year, as they race to secure financing to meet soaring demand for AI capacity  —  Taiwanese tech firms have completed a record $14.5 billion of debt deals so far this year, as they race to secure financing …

Reuters: Companies are using AI at their global capability centers in India to bring more creative work in-house, cutting turnaround times and reliance on ad agencies  —  Global companies are using AI at their Indian hubs to bring more creative work in-house, cutting turnaround time and their reliance …

Discussion | Link

Trump accounts will be similar to a traditional IRA. Kyle Mazza/Anadolu via Getty Images Trump Accounts will launch in July 2026, aimed at kids' investment savings. The government will give $1,000 to babies born between 2025 and 2028. Parents can open accounts through the IRS and make annual contributions. American parents will soon have another option to fund their kids' financial futures. "Trump Accounts," which were set up as part of the sweeping GOP-backed tax and spending law last year, will officially launch in July 2026 as an investment savings option for children. The accounts are set to go live on Thursday, and parents who have already enrolled in the accounts can start activating them, per a Wednesday exclusive from The Wall Street Journal. But people will only be able to start depositing funds in the accounts from July. Business Insider is breaking down everything you need to know about the new accounts — and how to open one. What is a Trump Account? Trump Accounts are investment accounts for kids. The federal government will contribute $1,000 for newborns to kick-start the account. Parents or guardians can then add their own money — up to $5,000 a year — to maximize growth, but contributing isn't required. Adults will have control of the account until their child turns 18. At age 18, the child can either withdraw the funds or continue to use the account for long-term investment. The account will be fully in the child's name. Who is eligible? Parents can open a Trump Account for any child under 18 with a valid Social Security number. Only babies born between January 1, 2025 and December 31, 2028 will receive a $1,000 contribution from the government. Dell Technologies founder Michael Dell and his wife Susan announced that they would fund an additional $250 to the accounts for children under 10 in areas with a median household income under $150,000. How does it work? The US Treasury will automatically make a one-time $1,000 deposit into each eligible child's account upon opening. Parents, friends, or employers can then add up to $5,000 per kid each year to the account starting on July 4, 2026. Families with multiple children can open multiple accounts. Visa has also announced it will allow credit card holders to add rewards into their child's account. The money can be invested in any diversified portfolio of low-cost index funds, like those that track the S&P 500. Account holders will be able to see the live account balance and stock performance through an app. Teenagers and young adults can contribute their own income once they start earning money. Having a Trump Account doesn't restrict them from saving in other accounts, too, like a Roth IRA. How do I open a Trump Account? Parents or guardians can open a Trump Account through the Internal Revenue Service. They will need to complete the newly-created IRS Form 4547 when filing their taxes this spring. The form asks for basic identifying details, such as a child's name, address, birthdate, Social Security number, and citizenship status. Parents will also be able to open an account for their child through an online government portal, which is set to launch over the summer. Updated information will be posted on trumpaccounts.gov How and when can I use the money? Trump Accounts will be locked from withdrawal until the calendar year a child turns 18, at which point the funds will be transferred to their control. Account holders can choose to use the money toward a down payment, college tuition, a new business, or other major purchases. This differs from other child-focused savings accounts like a 529, which can only be used for specific purposes, like higher education. Account holders don't have to withdraw the money at 18. They can continue to add money and treat it like a typical investment account into adulthood. How will the account be taxed? Trump Accounts will offer tax advantages similar to those of a traditional IRA. The $1,000 government-funded seed money won't be taxed as family income, and any parent contributions will be tax-deferred. These contributions must be made with after-tax dollars. Account holders won't owe taxes on investment earnings while they remain in the account, but the money will be taxed at ordinary income rates upon withdrawal. Can I contribute to Trump Accounts for my employees? If you're a business owner, you can contribute up to $2,500 annually to Trump Accounts on behalf of your employees' children. These deposits are tax-deductible and count toward the $5,000 annual contribution limit for the account. JP Morgan and Bank of America have already pledged to match $1,000 contributions to their employees' Trump accounts. Read the original article on Business Insider

Salesforce CEO Marc Benioff says AI can now read Slack chats in the workplace to identify what employees are upset about. Chris J. Ratcliffe/Bloomberg via Getty Images Marc Benioff said his company's engineering headcount has "been mostly flat" over the past two years. Benioff said the company is still growing its headcount, but in the sales department. Benioff said that AI agents fall short in selling and communicating. The hottest new hires are no longer engineers — at least not at Salesforce. Marc Benioff, the CEO of Salesforce, said during Wednesday's quarterly earnings call that the company's number of engineers has "been mostly flat" over the past two years at around 15,000, citing "new coding agents" with "dramatic capabilities." But talking about Salesforce's growing head count, Benioff said, "it's mostly growing in Miguel's area, in sales," referring to Miguel Milano, Salesforce's president and chief revenue officer. "Because, I think we all realize the one thing that we are doing here with you, selling and communicating, the agents are not exactly doing that," he said, adding that sales will be a critical arm of the company's expansion. While Salesforce's sales representatives are safe, others are not. Benioff said in September that Salesforce had cut about 4,000 support roles, which AI agents would handle. He's also been big on AI use within the company, saying earlier this month in an interview that Salesforce is projected to spend $300 million on Anthropic tokens on projects this year. Benioff's comments come as Silicon Valley companies are slashing headcount because of AI. Cloudflare cut roughly 20% of its workforce in May, citing "massive productivity gains" from AI usage, CEO Matthew Prince said during the company's first-quarter earnings call. The fintech company Block slashed nearly half of its workforce earlier in the year, with CEO Jack Dorsey saying in a shareholders' letter that "intelligence tools have changed what it means to build and run a company." Standard Chartered CEO Bill Winters went viral last week for saying he was replacing "lower-value human capital" — a comment he's since apologized for. But executives have said that some jobs are not yet capable of being replaced by AI. Duolingo CEO Luis von Ahn said in a podcast interview this month that AI can't match the work of his artists and designers, and he won't lower the quality of his app just to use AI. Benjamin Todd, the president of the career research nonprofit 80,000 Hours, told Business Insider that many key tasks in jobs aren't automatable, and that workers should focus on "safe skills" rather than "safe jobs." Read the original article on Business Insider

arXiv:2605.27377v1 Announce Type: new Abstract: We present RAG-Coding, an agentic method for automated ICD-10-CM coding. RAG-Coding orchestrates four large language model (LLM) agents and grounds their coding decisions in external knowledge sources (e.g. the official coding tabular list and guidelines). By retrieving and cross-referencing relevant knowledge in these sources, the agents enhance coding accuracy and ensure clinical compliance. On the MDACE dataset, RAG-Coding outperforms the best LLM-based baseline by 8-13\% in micro-F1 and 2-8\% in macro-F1 across multiple LLM backbones. Compared to the state-of-the-art pretrained language model method, PLM-ICD, RAG-Coding exhibits higher micro recall (+11\%), while PLM-ICD exhibits higher micro precision (+6\%), yielding comparable micro- and macro-F1. Ablations show stepwise gains, highlighting the importance of incorporating external knowledge. We also release MDACE-2025, updating the original dataset with expert re-annotations with the latest 2025 ICD-10-CM guidelines. This update features more fine-grained code labels and enables evaluation against current clinical standards.

arXiv:2605.27460v1 Announce Type: new Abstract: Single-frame atmospheric turbulence mitigation is inherently ill-posed due to spatially varying blur coupled with non-rigid geometric distortion. Existing end-to-end approaches trained on flat-field simulations often struggle to balance texture recovery with geometric rectification. To overcome this limitation, we propose D$^2$Turb, a unified framework that bridges physics-grounded simulation with explicitly decoupled restoration. First, we introduce a Depth-Aware Turbulence Synthesis protocol that incorporates scene depth into the phase-to-space formulation. This generates physically consistent, depth-dependent degradations and provides a crucial intermediate tilt supervision signal for disentangled learning. Building upon this simulation engine, D$^2$Turb decomposes restoration into two interactive stages: texture deblurring and geometric rectification. The texture deblurring stage employs a deblurring backbone to recover fine-grained details while preserving geometric distortion for the subsequent rectification stage. To mitigate the information fragmentation commonly observed in cascaded designs, we further propose an Adaptive Structural Prior Injection (ASPI) mechanism that dynamically transfers deep structural representations from the deblurring module to guide dense flow prediction for spatial unwarping. Extensive experiments demonstrate that D$^2$Turb achieves state-of-the-art performance on both synthetic and real-world datasets, with consistent improvements in both texture recovery and geometric fidelity. Our code and pre-trained models are publicly available at https://github.com/HertzDot222/D2Turb.

arXiv:2605.27465v1 Announce Type: new Abstract: The quadratic cost of self-attention in Vision Transformers (ViTs) constitutes a fundamental bottleneck for practical deployment, motivating a vibrant line of research on token reduction. Among existing approaches, token merging (ToMe) has emerged as an elegant training-free solution; yet its design rests on an unspoken premise of token equality, which contravenes the well-documented non-uniformity of self-attention and leads to information loss in high-salience tokens under aggressive compression. We address this limitation with AdaMerge, a token-merging framework based on two complementary mechanisms. First, salience-weighted similarity leverages column-wise feature-affinity centrality as a token-importance proxy and incorporates the resulting salience scores into the bipartite matching score, ensuring that pivotal tokens contribute more strongly to the merged representation. Second, adaptive merging intensity uses pre-computed layer-wise similarity statistics to dynamically modulate the per-layer reduction count in accordance with input-specific redundancy. On ImageNet-1k with ViT-B/16, AdaMerge consistently outperforms ToMe, PiToMe, and DSM across all FLOPs-matched regimes. The accuracy gap widens monotonically with compression: at the 13.4G FLOPs operating point, AdaMerge sustains a Top-1 degradation of only -1.06%, compared to -1.45% for PiToMe and -4.62% for DSM. To our knowledge, AdaMerge is the first to combine salience-weighted similarity and adaptive per-layer reduction into a single training-free token merging framework, advancing the accuracy-FLOPs Pareto frontier of ViT acceleration.

arXiv:2605.27387v1 Announce Type: new Abstract: Diffusion models promise efficient parallel text generation but rely on bidirectional attention, creating a structural mismatch with pre-trained Autoregressive (AR) models. This incompatibility precludes reusing robust AR priors, necessitating prohibitive pre-training from scratch. To bridge this gap, we propose FLUID, a framework that efficiently adapts AR backbones to the diffusion paradigm. By enforcing Strictly Causal Alignment, FLUID enables seamless initialization from standard GPT-style checkpoints, circumventing the need for massive pre-training. Furthermore, we introduce Elastic Horizons, an entropy-driven mechanism that dynamically modulates denoising strides based on local information density rather than fixed schedules. Experiments demonstrate that FLUID achieves state-of-the-art performance while reducing training costs by orders of magnitude, effectively reconciling established AR foundations with efficient parallel generation. Our code is available at https://github.com/Oli-lab-nun/FLUID/tree/main.

arXiv:2605.27378v1 Announce Type: new Abstract: Dental image analysis plays a pivotal role in supporting accurate diagnosis and treatment planning in oral healthcare. Although recent advances have produced dental AI models for specific tasks and individual imaging modalities, their isolated designs limit practical use in real-world clinical workflows. In this paper, we present OralAgent, the first dental-specialized AI agent that unifies multimodal reasoning, tool-based decision-making, and knowledge-grounded retrieval within an end-to-end automated framework. It integrates 22 visual analysis tools and 368 widely-used classical dental textbooks, enabling autonomous reasoning, planning, tool use, knowledge retrieval, and multi-step workflow execution. Furthermore, we introduce OralCorpus, a large-scale, high-quality bilingual textual resource containing 134.8M tokens curated for dental retrieval-augmented generation (RAG). To evaluate models' multidisciplinary dental knowledge, we construct OralQA-ZH, a Chinese multiple-choice question benchmark consisting of 798 items across eleven oral subspecialties. Extensive experiments demonstrate that OralAgent achieves state-of-the-art performance on the MMOral-Uni, MMOral-OPG, and OralQA-ZH benchmarks, highlighting its effectiveness, interpretability, and adaptability in real-world clinical settings. The code and models are publicly available at https://github.com/isjinghao/OralAgent.

arXiv:2605.27458v1 Announce Type: new Abstract: Transformer has significantly propelled the development of artificial intelligence, and certainly the development of agents as well. We categorize attention structures of Transformer into two types based on the source of the input information: homogenous and heterogenous attention structures. Heterogenous attention structures, with co-attention as a typical example, process information from different sources. Heterogenous attention structure is the foundation for Transformer models to achieve more complex functions and integrate more modal information. Whether for research purposes or policy requirements, the interpretation of Transformer models with heterogenous attention structures is an important task. The fusion of information from different sources brings new challenges. Our work mainly includes two parts: method and experimentation. In terms of method, we propose an interpretation method for Transformer models with heterogenous attention structures. In terms of experimentation, based on our experimental analysis paradigm, we interpret the operating mechanisms of representative models, conduct semantic interpretation and logical interpretation.

arXiv:2605.27385v1 Announce Type: new Abstract: Federated reinforcement learning (FedRL) enables multiple agents to collaboratively train a global policy without sharing raw data, making it ideal for privacy-sensitive applications. However, FedRL faces challenges in heterogeneous environments where differing state-transition dynamics lead to non-identical input distributions and imbalanced parameter updates during aggregation. Therefore, this paper develops a personalized observation normalization (PON) method, allowing each agent to locally normalize raw state inputs using a continuously updated running mean and variance. This design ensures consistent scaling of local feature without overshadowing across agents during aggregation. Furthermore, we demonstrate that sharing normalization parameters across agents is ineffective due to the diverse local input distributions, which highlights the necessity of personalized statistics. Experiments on heterogeneous MuJoCo tasks show that our developed PON accelerates training and achieves superior performance compared to baseline methods.

arXiv:2605.27469v1 Announce Type: new Abstract: Continual Learning (CL) is a practical paradigm to utilize power of deep pre-trained neural networks, but which pre-trained model has a better ability to balance ``Plasticity-Stability", deserving to be chosen? The logit shift serves as a natural proxy because it represents the logit shift in CL scenarios. However, obtaining the logit shift requires huge computational cost, which hinders large-scale model selection. Existing theoretical analyses fail to offer an efficient alternative because of the assumption of uniform hidden layer widths, which ignores the structural heterogeneity (variable width and depth) of real-world architectures. This raises a critical question: what theoretically relationship can be identified between heterogeneous architecture and logit shift on prior tasks (that the model has been trained on)? To answer the question, we decouple logit shift into architecture dependency and data dependency to establish our framework, which reveals that the combination of two dependency, defined as Architecture-driven Shift (ADS), that can capture the logit shift tendency well computable with few data samples. Specifically, for a well-optimized model on prior tasks, higher ADS is associated with a larger logit shift after training on the current task, which derived based on three mechanistic components: (1) spectral norm scaling of weight matrix gradients with layer width, (2) the optimization path length of the new task, and (3) the asymptotic task conflict in wide networks. Extensive empirical results across more than 175 diverse architectures demonstrate a strong monotonic correlation (the weakest Spearman's $r_s=0.731$) between ADS and logit shift. Practically, we demonstrate that ADS can serve as a lightweight proxy of the expected calibration error, which is a widely used metric for reliable CL model selection, on three datasets across six scenarios.

arXiv:2605.27374v1 Announce Type: new Abstract: Recent advances in multimodal large language models (MLLMs) and diffusion models (DMs) have opened new possibilities for AI-generated content. Yet, personalized cover image generation remains underexplored, despite its critical role in boosting user engagement on digital platforms. We propose ICG, a novel framework that integrates MLLM-based prompting with personalized preference alignment to generate high-quality, contextually relevant covers. ICG extracts semantic features from item titles and reference images via meta tokens, refines them with user embeddings, and injects the resulting personalized context into the diffusion model. To address the lack of labeled supervision, we adopt a multi-reward learning strategy that combines public aesthetic and relevance rewards with a personalized preference model trained from user behavior. Unlike prior pipelines relying on handcrafted prompts and disjointed modules, ICG employs an adapter to bridge MLLMs and diffusion models for end-to-end training. Experiments demonstrate that ICG significantly improves image quality, semantic fidelity, and personalization, leading to stronger user appeal and offline recommendation accuracy in downstream tasks. As a plug-and-play adapter bridging MLLMs and diffusion models, ICG is compatible with common checkpoints and requires no ground-truth labels during optimization.

arXiv:2605.27456v1 Announce Type: new Abstract: Geometric deep learning organises neural architectures around the symmetries of their data domain, with the choice of symmetry group serving as a geometric prior that determines what representations can be learned. Metric-Aware Principal Component Analysis (MAPCA) parameterises principal component analysis by a positive-definite metric matrix, with a canonical subfamily interpolating between standard PCA and output whitening and a diagonal-metric point recovering Invariant PCA (IPCA). This paper positions MAPCA within the geometric deep learning framework. The metric is read as the geometric prior; the orthogonal group preserving it is the symmetry group it induces; MAPCA solutions are equivariant under this group with the resulting spectrum invariant; and MAPCA's defining constraint is the linear analogue of the Schur-type weight constraints used in equivariant networks. Across six axes - domain, symmetry group, equivariance, invariance, architectural primitive, and geometric prior - we construct a precise dictionary between MAPCA and geometric deep learning. The technical anchor is a uniqueness theorem characterising IPCA as the unique linear data-derived metric in the MAPCA family that is equivariant under arbitrary diagonal rescaling and projects onto the fixed-point set of the action, equivalent under normalisation to the variance-maximisation criterion in its precise form. The paper closes with three bridges: kernel PCA as the nonlinear extension, spectral graph methods as MAPCA on graphs, and a deep MAPCA construction extending the positioning into deep equivariant networks

arXiv:2605.27428v1 Announce Type: new Abstract: Edge deployments of generative inference increasingly face two practical realities: per-device per-model performance is often unknown at deployment time, and it is non-stationary due to user-driven semantic events, background load, and device churn. Consequently, a resource manager that is tuned offline under a fixed regime can become brittle and expensive to maintain. This paper presents $E^3$-Agent, an executable and evolving agent for edge artificial intelligence generated content (AIGC) resource management. $E^3$-Agent separates a fast-path router that makes millisecond-level dispatch decisions from a slow-path, event-driven large language model (LLM) meta-controller that mitigates regime shifts through a small, explicit control surface exposed via a tool interface, including risk gating, router configuration, and rapid performance calibration. The agent learns online from execution feedback and continuously adapts to unknown and time-varying service-time mappings. We evaluate $E^3$-Agent in a discrete-event simulator driven by MLPerf-derived device-model measurement priors, covering cold-start warmup and three dynamic regimes: semantic dynamics, device churn, and hidden drift. Across the dynamic scenarios, $E^3$-Agent reduces average latency by 65%-73% compared to the best static baseline, stays within 7%-10% of an online full-information Oracle used for evaluation, and effectively suppresses stutter rate under semantic degradation.

arXiv:2605.27431v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) presents a naturally compatible and scalable framework for multimodal learning, demonstrating strong adaptability across diverse modalities and tasks. Despite its growing success, a comprehensive and systematic review on the MoE metho addressing multimodal challenges remains lacking. Existing surveys tend to evaluate either multimodal learning or MoE independently from method taxonomy, overlooking the unique interplay between them. This survey fills that gap by answering a central question: \textit{How does MoE effectively resolve multimodal challenges?} We approach this from three key perspectives: (1) \textbf{MoE as an Efficient Multimodal Engine:} enabling scalable multimodal modeling by decoupling computational cost from parameter growth and mitigating modality redundancy through selective expert activation; (2) \textbf{MoE as a Multimodal Representation Learner:} integrating complementary multi-opinion expert knowledge to enrich alignment and interaction representations; and (3) \textbf{MoE as a Multimodal Adapter:} providing a modular and flexible mechanism to model imperfect data scenarios such as modality imbalance and missing modality. Through our extensive literature review, we identify critical research gaps, including interpretable routing, expert communication, modality integration, and lifelong multimodal learning. We position this survey as a foundation for future research toward interpretable and sustainable multimodal Mixture-of-Experts system.