Blog
Advancing Embodied Agent Security: From Safety Benchmarks to Input Moderation
Embodied agents exhibit immense potential across a multitude of domains, making the assurance of their behavioral safety a fundamental prerequisite for their widespread deployment. However, existing research predominantly concentrates on the security of general large language models, lacking specialized methodologies for establishing...
Spin and energy diffusion vs. subdiffusion in disordered spin chains
While the high-temperature spin diffusion in spin chains with random local fields has been the subject of numerous studies concerning the phenomenon of many-body localization (MBL), the energy diffusion in the same models has been much less explored. We show that energy...
Riesz transform, function spaces and their applications on infinite dimensional compact groups
On a compact connected group $G$, consider the infinitesimal generator $-L$ of a central symmetric Gaussian convolution semigroup $(\mu_t)_{t>0}$. We establish several regularity results of the solution to the Poisson equation $LU=F$, both in strong and weak senses. To this end, we...
SeaLLM: Service-Aware and Latency-Optimized Resource Sharing for Large Language Model Inference
Large language models (LLMs) with different architectures and sizes have been developed. Serving each LLM with dedicated GPUs leads to resource waste and service inefficiency due to the varying demand of LLM requests. A common practice is to share multiple LLMs. However,...
From predictions to confidence intervals: an empirical study of conformal prediction methods for in-context learning
Transformers have become a standard architecture in machine learning, demonstrating strong in-context learning (ICL) abilities that allow them to learn from the prompt at inference time. However, uncertainty quantification for ICL remains an open challenge, particularly in noisy regression tasks. This paper...
Low-overhead error detection with spacetime codes
We introduce a low-overhead approach for detecting errors in arbitrary Clifford circuits on arbitrary qubit connectivities. Our method is based on the framework of spacetime codes, and is particularly suited to near-term hardware since it has a much milder overhead in qubits...
SAGA: Semantic-Aware Gray color Augmentation for Visible-to-Thermal Domain Adaptation across Multi-View Drone and Ground-Based Vision Systems
Domain-adaptive thermal object detection plays a key role in facilitating visible (RGB)-to-thermal (AND) adaptation by reducing the need for co-registered image pairs and minimizing reliance on large annotated IR datasets. However, inherent limitations of IR images, such as the lack of color...
Enhancing Radiation Hardness and Granularity in HV-CMOS: The RD50-MPW4 Sensor
The latest HV-CMOS pixel sensor developed by the former CERN-RD50-CMOS group, known as the \mpw, demonstrates competitive radiation tolerance, spatial granularity, and timing resolution — key requirements for future high-energy physics experiments such as the HL-LHC and FCC. Fabricated using a \SI{150}{nm}...
Riemannian Neural Geodesic Interpolant
Stochastic interpolants are efficient generative models that bridge two arbitrary probability density functions in finite time, enabling flexible generation from the source to the target distribution or vice versa. These models are primarily developed in Euclidean space, and are therefore limited in...
Development of an Ultra-fast, Likelihood-based, Distance Inference Framework for the Next Generation of Type Ia Supernova Surveys
In this work, we present EDRIS (French for Distance Estimator for Incomplete Supernova Surveys), a cosmological inference framework tailored to reconstruct unbiased cosmological distances from type Ia supernovae light-curve parameters. This goal is achieved by including data truncation directly in the statistical...