Post-training alignment determines the reasoning and human preference following capabilities of large language models, yet most existing works withhold detailed data construction, filtering rules and training recipes, which hinders community reproducibility and lightweight model optimization. This work presents NebulaExp, a fully transparent, ablation-driven post-training pipeline built on Qwen3-8B-base, covering two orthogonal model branches: general instruct model and complex reasoning-specialized model. We curate a raw corpus of 3.84M multi-source SFT samples and a 200K verifiable RL candidate pool, and design an end-to-end data processing stack including response distillation, multi-dimensional cross-verification filtering, fine-grained difficulty grading, task classification and diversity-aware sampling. For the Instruct branch, our three-stage optimized supervised fine-tuning approach NebulaExp-Ins-SFT improves the average benchmark score from the 55.01 baseline of Qwen3-8B-nothink to 60.99. GRPO reinforcement learning then further elevates the average score to 61.85. For the Reasoning branch, medium-difficulty GRPO RL improves average reasoning score from 73.88 to 75.17. To
This paper presents HITSZ's submission for the IWSLT 2025 Indic track, focusing on speech-to-text translation (ST) for English-to-Indic and Indic-to-English language pairs. To enhance translation quality in this low-resource scenario, we propose an end-to-end system integrating the pre-trained Whisper automated speech recognition (ASR) model with Krutrim, an Indic-specialized large language model (LLM). Experimental results demonstrate that our end-to-end system achieved average BLEU scores of $28.88$ for English-to-Indic directions and $27.86$ for Indic-to-English directions. Furthermore, we investigated the Chain-of-Thought (CoT) method. While this method showed potential for significant translation quality improvements on successfully parsed outputs (e.g. a $13.84$ BLEU increase for Tamil-to-English), we observed challenges in ensuring the model consistently adheres to the required CoT output format.
The transferability of adversarial examples poses a significant security challenge for deep neural networks, which can be attacked without knowing anything about them. In this paper, we propose a new Segmented Gaussian Pyramid (SGP) attack method to enhance the transferability, particularly against defense models. Unlike existing methods that generally focus on single-scale images, our approach employs Gaussian filtering and three types of downsampling to construct a series of multi-scale examples. Then, the gradients of the loss function with respect to each scale are computed, and their average is used to determine the adversarial perturbations. The proposed SGP can be considered an input transformation with high extensibility that is easily integrated into most existing adversarial attacks. Extensive experiments demonstrate that in contrast to the state-of-the-art methods, SGP significantly enhances attack success rates against black-box defense models, with average attack success rates increasing by 2.3% to 32.6%, based only on transferability.
Adversarial attacks have become a significant challenge in the security of machine learning models, particularly in the context of black-box defense strategies. Existing methods for enhancing adversarial transferability primarily focus on the spatial domain. This paper presents Frequency-Space Attack (FSA), a new adversarial attack framework that effectively integrates frequency-domain and spatial-domain transformations. FSA combines two key techniques: (1) High-Frequency Augmentation, which applies Fourier transform with frequency-selective amplification to diversify inputs and emphasize the critical role of high-frequency components in adversarial attacks, and (2) Hierarchical-Gradient Fusion, which merges multi-scale gradient decomposition and fusion to capture both global structures and fine-grained details, resulting in smoother perturbations. Our experiment demonstrates that FSA consistently outperforms state-of-the-art methods across various black-box models. Notably, our proposed FSA achieves an average attack success rate increase of 23.6% compared with BSR (CVPR 2024) on eight black-box defense models.
Large Language Models (LLMs) have demonstrated remarkable instruction-following capabilities across various applications. However, their performance in multilingual settings lacks systematic investigation, with existing evaluations lacking fine-grained constraint analysis across diverse linguistic contexts. We introduce XIFBench, a comprehensive constraint-based benchmark for evaluating multilingual instruction-following abilities of LLMs, comprising 558 instructions with 0-5 additional constraints across five categories (Content, Style, Situation, Format, and Numerical) in six languages spanning different resource levels. To support reliable and consistent cross-lingual evaluation, we implement three methodological innovations: cultural accessibility annotation, constraint-level translation validation, and requirement-based evaluation using English requirements as semantic anchors across languages. Extensive experiments with various LLMs not only quantify performance disparities across resource levels but also provide detailed insights into how language resources, constraint categories, instruction complexity, and cultural specificity influence multilingual instruction-following.
Nitrogen-vacancy (NV) centers show great potentials for nanoscale bio-sensing and bio-imaging. Nevertheless, their envisioned bio-applications suffer from intrinsic background noise due to unavoidable light scattering and autofluorescence in cells and tissues. Herein, we develop a novel all-optical modulated imaging method via physically-enabled classifier, for on-demand and direct access to NV fluorescence at pixel resolution while effectively filtering out background noise. Specifically, NV fluorescence can be modulated optically to exhibit sinusoid-like variations, providing basis for classification. We validate our method in various complex biological scenarios with fluorescence interference, ranging from cells to organisms. Notably, our classification-based approach achieves almost 10^6 times enhancement of signal-to-background ratio (SBR) for fluorescent nanodiamonds (FNDs) in neural protein imaging. We also demonstrate 4-fold contrast improvement in optically-detected magnetic resonance measurements (ODMR) of FNDs inside stained cells. Our technique offers a generic, explainable and robust solution, applicable for realistic high-fidelity imaging and sensing in challenging no
The contactless temperature measurement at micro/nanoscale is vital to a broad range of fields in modern science and technology. The nitrogen vacancy (NV) center, a kind of diamond defect with unique spin-dependent photoluminescence, has been recognized as one of the most promising nanothermometers. However, this quantum thermometry technique has been prone to a number of possible perturbations, which will unavoidably degrade its actual temperature sensitivity. Here, for the first time, we have developed a cross-validated optical thermometry method using a bulk diamond sample containing both NV centers and silicon vacancy (SiV) centers. Particularly, the latter allowing all-optical method has been intrinsically immune to those influencing perturbations for the NV-based quantum thermometry, hence serving as a real-time cross validation system. As a proof-of-concept demonstration, we have shown a trustworthy temperature measurement under the influence of varying magnetic fields. This multi-modality approach allows a synchronized cross-validation of the measured temperature, which is required for micro/nanoscale quantum thermometry in complicated environments such as a living cell.
Machine learning algorithms have been proven effective for essential quantum computation tasks such as quantum error correction and quantum control. Efficient hardware implementation of these algorithms at cryogenic temperatures is essential. Here, we utilize magnetic topological insulators as memristors (termed magnetic topological memristors) and introduce a cryogenic in-memory computing scheme based on the coexistence of the chiral edge state and the topological surface state. The memristive switching and reading of the giant anomalous Hall effect exhibit high energy efficiency, high stability, and low stochasticity. We achieve high accuracy in a proof-of-concept classification task using four magnetic topological memristors. Furthermore, our algorithm-level and circuit-level simulations of large-scale neural networks demonstrate software-level accuracy and lower energy consumption for image recognition and quantum state preparation compared with existing magnetic memristor and CMOS technologies. Our results not only showcase a new application of chiral edge states but also may inspire further topological quantum physics-based novel computing schemes.
Hidden moon ice may reveal itself through the way it bends and reflects vibrations from moonquakes。 The technique could help astronauts locate vital water supplies and uncover clues about the origins of Earth’s oceans
A surprisingly simple change could make sodium-ion batteries far more powerful while opening the door to turning seawater into drinking water。 Researchers at the University of Surrey found that sodium vanadium oxide performs much better when its naturally occurring water is left inside instead of being removed during manufacturing
Scientists traced a mysterious surge of low-energy gamma rays from zinc-70 to magnetic changes occurring inside its nucleus。 The breakthrough could improve models of how stars, supernovae, and neutron star mergers create heavy elements
The solar wind may be stripping away Mars’ atmosphere by creating giant rolling waves along its outer edge。 Data from MAVEN and Tianwen-1 show that these waves form plasma clouds that carry atmospheric particles into space。 The process is concentrated on one side of Mars and may have played a major role in drying out the once potentially habitable
NASA’s Psyche spacecraft aced its Mars flyby, using the planet’s gravity to speed toward its 2029 encounter with a metal-rich asteroid。 The spacecraft tested its cameras, magnetometer, and particle-detecting instruments, capturing unusual views of Mars and measuring its magnetic environment。 It also detected neutrons from the planet and spotted the
Researchers have found a way to build much larger “twisted” oxide materials while precisely controlling how their atomic layers line up。 Because these materials can be made over large areas and transferred onto different surfaces, the technique could help turn twistronics from a laboratory curiosity into a practical platform for next-generation ele
NASA’s Swift Observatory observed a supermassive black hole ripping apart a star more than 30,000 light-years from the center of a distant galaxy。 The extraordinary flare briefly outshone its entire host galaxy in ultraviolet light and revealed a black hole about a million times the Sun’s mass
Primordial black holes may occasionally pass through white dwarf stars and trigger enormous Type Ia supernova explosions。 Researchers found that these events could explain chemical patterns seen in supernova remnants, nearby explosions, and stars across the Milky Way
Scientists have uncovered new evidence that Venus may still be tearing itself apart from within。 Advanced 3D simulations indicate that some of the planet's giant rift valleys formed relatively recently and could still be expanding。 The results suggest Venus has a far more active interior than researchers once believed, challenging the long-held vie
Physicists have uncovered a surprising limit to electrical resistance caused by particles colliding。 Using ultracold potassium atoms trapped in a grid of light, researchers created a highly controlled stand-in for electrons moving through a solid。 As collisions became more frequent and intense, resistance initially rose, but eventually hit a ceilin