2026年10月6日

Samsung secures customer backing for US$5B substrate expansion

Samsung Electro-Mechanics is investing KRW6.78 trillion (approx. US$5 billion) to expand FCBGA capacity in South Korea and Vietnam. The projects are backed by customer funding and medium- to long-term volume commitments,…

原文链接

Taiwan bets on SiPh, CPO to tackle AI data center power bottlenecks

Taiwan's National Science and Technology Council (NSTC) has launched two research programs targeting silicon photonics (SiPh) and next-generation AI data center power systems, seeking to address energy and performance bo…

原文链接

TSMC tops global FDI ranking as AI infrastructure redraws overseas investment

The direction of global corporate investment in physical assets is becoming clearer. According to the fDi Report 2026 and an analysis by Visual Capitalist, announced greenfield foreign direct investment by multinational …

原文链接

Taiwan's NSTC seeks AI data center power tech to reduce grid impact of load spikes

Taiwan's National Science and Technology Council (NSTC) has recently issued a call for proposals on "high-performance power conversion and system management technologies for next-generation AI data centers." The aim is t…

原文链接

Apple, Qualcomm renew patent deal, iPhone modem shift deepens

Qualcomm announced on September 24 that it has renewed its global patent licensing agreement with Apple, with the new contract set to take effect on April 1, 2027. The terms and value of the deal were not disclosed, but …

原文链接

Drone share tops 10% as Calin Technology eyes LEO optics, auto growth

Optical lens maker Calin Technology is seeing its new business ventures bear fruit. Revenue from its drone camera module segment has surpassed 10% of total sales in 2026 and is targeting over 15%, making it the company's…

原文链接

Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice Text-to-Speech Model

We distill Kokoro-82M, a widely used open text-to-speech model with 54 voices, into Paradee, an 8.07M-parameter model that speaks one of them. Paradee keeps Kokoro's architecture with much narrower layers, and each of it…

原文链接

Private online learning and prediction for Littlestone classes

We study mistake bounds for differentially private online learning and online prediction under oblivious realisable adversaries. Online learning requires the learner to release a hypothesis at each time step whereas in o…

原文链接

Deep Learning for Sleep Heart Rate Estimation from Accelerometers: Toward Population-Scale Cardiac Insight Without Optic

Large longitudinal cohorts often contain wrist accelerometry without optical heart-rate sensing, motivating recovery of cardiac information from motion signals already collected during sleep. We present SeqSmoother, a tr…

原文链接

TasteVal: Measuring the Experimental Research Taste of AI Systems Against Human Experts

We introduce TasteVal, a benchmark to evaluate the experimental research taste of frontier models. We define research taste as the ability to pick interesting problems to solve, design experiments, and interpret experime…

原文链接

CLIFT: Conformal Self-Verification for Web Agent Training and Test-Time Scaling

Open-source web agents are now strong enough to execute realistic browser tasks, but training them with reinforcement learning still depends on weak supervision: binary task success is too sparse for credit assignment, w…

原文链接

MemPilot: Orchestrating On-Demand Multimodal Memory Curation for LLM Agents

Memory has become integral to the LLM agent ecosystem, supporting information retention and reuse across interactions. However, most existing agent memory systems construct memory in a query-agnostic manner, which can in…

原文链接

UniSlider: Perceptually Uniform Sliders for Continuous Image Editing

Sliders provide an intuitive interface for continuous image editing. In current generative approaches, however, the slider is simply a rescaling of the method's strength parameter, such as an adapter coefficient, a promp…

原文链接

Towards Looped Models Done Right, Part II: Rethinking at Fixed Points

Every recurrence of a looped language model adds cost in training, decoding, prefill, and reinforcement learning (RL). The closer recurrent states get to fixed points, the less the path to them matters. This enables trun…

原文链接

Direct Intermediate Initialization for Tilted Diffusion Samplers

Some diffusion posterior samplers construct Gaussian-tilted intermediate distributions along the reverse process. We observe that these targets can be pulled back to clean-space posteriors with weaker conditioning, with …

原文链接

Recursive Video In-Context Learning for Agentic Robot

LLM agents that orchestrate frozen vision-language-action (VLA) policies improve across episodes through text memory, which records what the agent did but not how the task is done. A demonstration video shows it, but fit…

原文链接

Learning to Read the Contextual Tokens in Diffusion Transformers

Multimodal Diffusion Transformers (MM-DiTs) jointly process visual and textual representations throughout generation. These models repeatedly update the text tokens through multimodal attention, forming dynamic contextua…

原文链接

BiasFlow: Geometric Monitoring and Backbone Regularization for Spurious Feature Reliance

Worst-group accuracy (WGA) evaluates a trained predictor but does not characterize how its frozen backbone behaves when a new head is learned. We introduce BiasFlow, a hook-based toolkit for monitoring class-attribute ce…

原文链接

InterMimicGen: Scaling Humanoid Loco-Manipulation through Self-Evolving Motion Imitation

Captured human-object interactions provide rich supervision for humanoid loco-manipulation, but they are sparse, heterogeneous, and not directly executable by robots. We introduce InterMimicGen, a self-evolving motion-im…

原文链接

Base Models Can Reason By Taking a Cue From Training Data

In this paper, we study how training data creates associations between the tokens at the start of a base model's response and the reasoning behavior that follows. First, we demonstrate that fixing particular starting tok…

原文链接

One Figure, Every Canvas: Editable Flowchart Relayout via Agentic Pipeline

Pipeline figures in ML papers must be repurposed across many canvases, including paper columns, 16:9 slides, portrait posters, 1:1 social teasers, 9:16 phone previews. Each format imposes a different aspect ratio on the …

原文链接

Sumitomo Corporation invests in Hero Future Energies’ solar-wind-storage hybrid in Karnataka, India

Japanese trading company Sumitomo Corporation has invested in a utility-scale renewable energy park combining solar PV, wind and energy storage in India. Developed by Indian independent power producer (IPP) Hero Future E…

原文链接

Australia’s NEM battery storage fleet earns AU$88.91 million in Q3, with five days delivering 15% of revenue

原文链接

Australia’s Origin Energy targets building ‘Southern Hemisphere’s largest battery storage complex’

原文链接

China National Day EV charging surges 49% as highway grid faces peak-load test

China's National Day holidays have driven a massive surge in long-distance electric vehicle (EV) travel. According to data from China's National Energy Administration cited by The Paper, the 62,700 monitored highway char…

原文链接