arXiv subjects

Sheng-Chieh Lin

Publications and source records attributed to Sheng-Chieh Lin.

At least 19 recordsLinked to original sources

arXiv2026

Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation

We introduce Nemotron-Cascade 2, an open 30B MoE model with 3B activated parameters that delivers best-in-class reasoning and strong agentic capabilities. Despite its compact size, its mathematical and coding reasoning performance approaches that of frontier open models. It is the second open-weight LLM, after DeepSeekV3.2-Speciale-671B-A37B, to achieve Gold Medal-level performance in the 2025 International Mathematical Olympiad (IMO), the International Olympiad in Informatics (IOI), and the ICPC World Finals, demonstrating remarkably high intelligence density with 20x fewer parameters. In contrast to Nemotron-Cascade 1, the key technical advancements are as follows. After SFT on a meticulously curated dataset, we substantially expand Cascade RL to cover a much broader spectrum of reasoning and agentic domains. Furthermore, we introduce multi-domain on-policy distillation from the strongest intermediate teacher models for each domain throughout the Cascade RL process, allowing us to efficiently recover benchmark regressions and sustain strong performance gains along the way. We release the collection of model checkpoint and training data.

cs.CL↗

arXiv2025

Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models

Building general-purpose reasoning models with reinforcement learning (RL) entails substantial cross-domain heterogeneity, including large variation in inference-time response lengths and verification latency. Such variability complicates the RL infrastructure, slows training, and makes training curriculum (e.g., response length extension) and hyperparameter selection challenging. In this work, we propose cascaded domain-wise reinforcement learning (Cascade RL) to develop Nemotron-Cascade, capable of operating in both instruct and deep thinking modes, without any performance gap relative to a thinking-only counterpart. Departing from conventional approaches that blend heterogeneous prompts from different domains, Cascade RL orchestrates sequential, domain-wise RL, reducing engineering complexity and delivering state-of-the-art performance across a wide range of benchmarks. Notably, RLHF for alignment, when used as a pre-step, boosts the model's reasoning ability far beyond mere preference optimization, and subsequent domain-wise RLVR stages rarely degrade the benchmark performance attained in earlier domains and may even improve it (see an illustration in Figure 1). Our 14B model, after RL, outperforms its SFT teacher, DeepSeek-R1-0528, on LiveCodeBench v5/v6/Pro and achieves silver-medal performance in the 2025 International Olympiad in Informatics (IOI). We transparently share our training and data recipes.

cs.CL↗

arXiv2025

Mapping the Nearest Ancient Sloshing Cold Front in the Sky with XMM-Newton

The Virgo Cluster is the nearest cool core cluster that features two well-studied sloshing cold fronts at radii of $r \approx 30$ kpc and $r \approx 90$ kpc, respectively. In this work, we present results of XMM-Newton mosaic observations of a third, southwestern, cold front at a radius of $r \approx 250$ kpc, originally discovered with Suzaku. All three cold fronts are likely to be parts of an enormous swirling pattern, rooted in the core. The comparison with a numerical simulation of a binary cluster merger indicates that these cold fronts were produced in the same single event $-$ likely the infall of M49 from the northwest of Virgo and it is now re-entering the cluster from the south. This outermost cold front has probably survived for $2-3$ Gyr since the disturbance. We identified single sharp edges in the surface brightness profiles of the southern and southwestern sections of the cold front, whereas the western section is better characterized with double edges. This implies that magnetic fields have preserved the leading edge of the cold front, while its western side is beginning to split into two cold fronts likely due to Kelvin-Helmholtz instabilities. The slopes of the 2D power spectrum of the X-ray surface brightness fluctuations, derived for the brighter side of the cold front, are consistent with the expectation from Kolmogorov turbulence. Our findings highlight the role of cold fronts in shaping the thermal dynamics of the intracluster medium beyond the cluster core, which has important implications for cluster cosmology. Next-generation X-ray observatories, such as the proposed AXIS mission, will be ideal for identifying and characterizing ancient cold fronts.

astro-ph.HE↗

arXiv2024

MM-Embed: Universal Multimodal Retrieval with Multimodal LLMs

State-of-the-art retrieval models typically address a straightforward search scenario, in which retrieval tasks are fixed (e.g., finding a passage to answer a specific question) and only a single modality is supported for both queries and retrieved results. This paper introduces techniques for advancing information retrieval with multimodal large language models (MLLMs), enabling a broader search scenario, termed universal multimodal retrieval, where multiple modalities and diverse retrieval tasks are accommodated. To this end, we first study fine-tuning an MLLM as a bi-encoder retriever on 10 datasets with 16 retrieval tasks. Our empirical results show that the fine-tuned MLLM retriever is capable of understanding challenging queries, composed of both text and image, but it underperforms compared to a smaller CLIP retriever in cross-modal retrieval tasks due to the modality bias exhibited by MLLMs. To address the issue, we propose modality-aware hard negative mining to mitigate the modality bias exhibited by MLLM retrievers. Second, we propose continuously fine-tuning the universal multimodal retriever to enhance its text retrieval capability while preserving multimodal retrieval capability. As a result, our model, MM-Embed, achieves state-of-the-art performance on the multimodal retrieval benchmark M-BEIR, which spans multiple domains and tasks, while also surpassing the state-of-the-art text retrieval model, NV-Embed-v1, on the MTEB retrieval benchmark. We also explore prompting the off-the-shelf MLLMs as zero-shot rerankers to refine the ranking of the candidates from the multimodal retriever. We find that, through prompt-and-reranking, MLLMs can further improve multimodal retrieval when the user queries (e.g., text-image composed queries) are more complex and challenging to understand. These findings also pave the way for advancing universal multimodal retrieval in the future.

cs.CL↗

arXiv2024

Semi-supervised Learning for Detecting Inverse Compton Emission in Galaxy Clusters

Inverse Compton (IC) emission associated with the non-thermal component of the intracluster medium (ICM) has been a long sought phenomenon in cluster physics. Traditional spectral fitting often suffers from the degeneracy between the two-temperature thermal spectrum (2T) and the one-temperature plus IC power-law spectrum (1T+IC). We present a semi-supervised deep learning approach to search for IC emission in galaxy clusters. We employ a conditional autoencoder (CAE), which is based on an autoencoder with latent representations trained to constrain the thermal parameters of the ICM. The algorithm is trained and tested using synthetic NuSTAR X-ray spectra with instrumental and astrophysical backgrounds included. The training data set only contains 2T spectra, which is more common than 1T+IC spectra. Anomaly detection is performed on the validation and test datasets, consisting of 2T spectra as the normal set and 1T+IC spectra as anomalies. With a threshold anomaly score, chosen based on cross-validation, our algorithm is able to identify spectra that contain an IC component in the test dataset, with a balanced accuracy (BAcc) of 0.64, which outperforms traditional spectral fitting (BAcc = 0.55) and ordinary autoencoder (BAcc = 0.55). Traditional spectral fitting is better at identifying IC cases among true IC spectra (a better recall), while IC predictions made by CAE have a higher chance of being true IC cases (a better precision), demonstrating their mutual complement to each other.

astro-ph.CO↗

arXiv2024

Synergistic Approach for Simultaneous Optimization of Monolingual, Cross-lingual, and Multilingual Information Retrieval

Information retrieval across different languages is an increasingly important challenge in natural language processing. Recent approaches based on multilingual pre-trained language models have achieved remarkable success, yet they often optimize for either monolingual, cross-lingual, or multilingual retrieval performance at the expense of others. This paper proposes a novel hybrid batch training strategy to simultaneously improve zero-shot retrieval performance across monolingual, cross-lingual, and multilingual settings while mitigating language bias. The approach fine-tunes multilingual language models using a mix of monolingual and cross-lingual question-answer pair batches sampled based on dataset size. Experiments on XQuAD-R, MLQA-R, and MIRACL benchmark datasets show that the proposed method consistently achieves comparable or superior results in zero-shot retrieval across various languages and retrieval tasks compared to monolingual-only or cross-lingual-only training. Hybrid batch training also substantially reduces language bias in multilingual retrieval compared to monolingual training. These results demonstrate the effectiveness of the proposed approach for learning language-agnostic representations that enable strong zero-shot retrieval performance across diverse languages.

cs.IR↗

arXiv2024

Unifying Multimodal Retrieval via Document Screenshot Embedding

In the real world, documents are organized in different formats and varied modalities. Traditional retrieval pipelines require tailored document parsing techniques and content extraction modules to prepare input for indexing. This process is tedious, prone to errors, and has information loss. To this end, we propose Document Screenshot Embedding (DSE), a novel retrieval paradigm that regards document screenshots as a unified input format, which does not require any content extraction preprocess and preserves all the information in a document (e.g., text, image and layout). DSE leverages a large vision-language model to directly encode document screenshots into dense representations for retrieval. To evaluate our method, we first craft the dataset of Wiki-SS, a 1.3M Wikipedia web page screenshots as the corpus to answer the questions from the Natural Questions dataset. In such a text-intensive document retrieval setting, DSE shows competitive effectiveness compared to other text retrieval methods relying on parsing. For example, DSE outperforms BM25 by 17 points in top-1 retrieval accuracy. Additionally, in a mixed-modality task of slide retrieval, DSE significantly outperforms OCR text retrieval methods by over 15 points in nDCG@10. These experiments show that DSE is an effective document retrieval paradigm for diverse types of documents. Model checkpoints, code, and Wiki-SS collection will be released.

cs.IR↗

arXiv2024

FLAME: Factuality-Aware Alignment for Large Language Models

Alignment is a standard procedure to fine-tune pre-trained large language models (LLMs) to follow natural language instructions and serve as helpful AI assistants. We have observed, however, that the conventional alignment process fails to enhance the factual accuracy of LLMs, and often leads to the generation of more false facts (i.e. hallucination). In this paper, we study how to make the LLM alignment process more factual, by first identifying factors that lead to hallucination in both alignment steps:\ supervised fine-tuning (SFT) and reinforcement learning (RL). In particular, we find that training the LLM on new knowledge or unfamiliar texts can encourage hallucination. This makes SFT less factual as it trains on human labeled data that may be novel to the LLM. Furthermore, reward functions used in standard RL can also encourage hallucination, because it guides the LLM to provide more helpful responses on a diverse set of instructions, often preferring longer and more detailed responses. Based on these observations, we propose factuality-aware alignment, comprised of factuality-aware SFT and factuality-aware RL through direct preference optimization. Experiments show that our proposed factuality-aware alignment guides LLMs to output more factual responses while maintaining instruction-following capability.

cs.CL↗

arXiv2023

Exploring chemical enrichment of the intracluster medium with the Line Emission Mapper

Synthesized in the cores of stars and supernovae, most metals disperse over cosmic scales and are ultimately deposited well outside the gravitational potential of their host galaxies. Since their presence is well visible through their X-ray emission lines in the hot gas pervading galaxy clusters, measuring metal abundances in the intracluster medium (ICM) offers us a unique view of chemical enrichment of the Universe as a whole. Despite extraordinary progress in the field thanks to four decades of X-ray spectroscopy using CCD (and gratings) instruments, understanding the precise stellar origins of the bulk of metals, and when the latter were mixed on Mpc scales, requires an X-ray mission capable of spatial, non-dispersive high resolution spectroscopy covering at least the soft X-ray band over a large field of view. In this White Paper, we demonstrate how the Line Emission Mapper (LEM) probe mission concept will revolutionize our current picture of the ICM enrichment. Specifically, we show that LEM will be able to (i) spatially map the distribution of ten key chemical elements out to the virial radius of a nearby relaxed cluster and (ii) measure metal abundances in serendipitously discovered high-redshift protoclusters. Altogether, these key observables will allow us to constrain the chemical history of the largest gravitationally bound structures of the Universe. They will also solve key questions such as the universality of the initial mass function (IMF) and the initial metallicity of the stellar populations producing these metals, as well as the relative contribution of asymptotic giant branch (AGB) stars, core-collapse, and Type Ia supernovae to enrich the cosmic web over Mpc scales. Concrete observing strategies are also briefly discussed.

astro-ph.GA↗

arXiv2023

Mapping the Intracluster Medium in the Era of High-resolution X-ray Spectroscopy

High-resolution spectroscopy in soft X-rays will open a new window to map multiphase gas in galaxy clusters and probe physics of the intracluster medium (ICM), including chemical enrichment histories, circulation of matter and energy during large-scale structure evolution, stellar and black hole feedback, halo virialization, and gas mixing processes. An eV-level spectral resolution, large field-of-view, and effective area are essential to separate cluster emissions from the Galactic foreground and efficiently map the cluster outskirts. Several mission concepts that meet these criteria have been proposed recently, e.g., LEM, HUBS, and SuperDIOS. This theoretical study explores what information on ICM physics could be recovered with such missions and the associated challenges. We emphasize the need for a comprehensive comparison between simulations and observations to interpret the high-resolution spectroscopic observations correctly. Using Line Emission Mapper (LEM) characteristics as an example, we demonstrate that it enables the use of soft X-ray emission lines (e.g., O VII/VIII and Fe-L complex) from the cluster outskirts to measure the thermodynamic, chemical, and kinematic properties of the gas up to $r_{200}$ and beyond. By generating mock observations with full backgrounds, analysing their images/spectra with observational approaches, and comparing the recovered characteristics with true ones from simulations, we develop six key science drivers for future missions, including the exploration of multiphase gas in galaxy clusters (e.g., temperature fluctuations, phase-space distributions), metallicity, ICM gas bulk motions and turbulence power spectra, ICM-cosmic filament interactions, and advances for cluster cosmology.

astro-ph.HE↗

arXiv2023

Aggretriever: A Simple Approach to Aggregate Textual Representations for Robust Dense Passage Retrieval

Pre-trained language models have been successful in many knowledge-intensive NLP tasks. However, recent work has shown that models such as BERT are not ``structurally ready'' to aggregate textual information into a [CLS] vector for dense passage retrieval (DPR). This ``lack of readiness'' results from the gap between language model pre-training and DPR fine-tuning. Previous solutions call for computationally expensive techniques such as hard negative mining, cross-encoder distillation, and further pre-training to learn a robust DPR model. In this work, we instead propose to fully exploit knowledge in a pre-trained language model for DPR by aggregating the contextualized token embeddings into a dense vector, which we call agg*. By concatenating vectors from the [CLS] token and agg*, our Aggretriever model substantially improves the effectiveness of dense retrieval models on both in-domain and zero-shot evaluations without introducing substantial training overhead. Code is available at https://github.com/castorini/dhr

cs.IR↗

arXiv2023

SLIM: Sparsified Late Interaction for Multi-Vector Retrieval with Inverted Indexes

This paper introduces Sparsified Late Interaction for Multi-vector (SLIM) retrieval with inverted indexes. Multi-vector retrieval methods have demonstrated their effectiveness on various retrieval datasets, and among them, ColBERT is the most established method based on the late interaction of contextualized token embeddings of pre-trained language models. However, efficient ColBERT implementations require complex engineering and cannot take advantage of off-the-shelf search libraries, impeding their practical use. To address this issue, SLIM first maps each contextualized token vector to a sparse, high-dimensional lexical space before performing late interaction between these sparse token embeddings. We then introduce an efficient two-stage retrieval architecture that includes inverted index retrieval followed by a score refinement module to approximate the sparsified late interaction, which is fully compatible with off-the-shelf lexical search libraries such as Lucene. SLIM achieves competitive accuracy on MS MARCO Passages and BEIR compared to ColBERT while being much smaller and faster on CPUs. To our knowledge, we are the first to explore using sparse token representations for multi-vector retrieval. Source code and data are integrated into the Pyserini IR toolkit.

cs.IR↗

arXiv2023

Improving Conversational Passage Re-ranking with View Ensemble

This paper presents ConvRerank, a conversational passage re-ranker that employs a newly developed pseudo-labeling approach. Our proposed view-ensemble method enhances the quality of pseudo-labeled data, thus improving the fine-tuning of ConvRerank. Our experimental evaluation on benchmark datasets shows that combining ConvRerank with a conversational dense retriever in a cascaded manner achieves a good balance between effectiveness and efficiency. Compared to baseline methods, our cascaded pipeline demonstrates lower latency and higher top-ranking effectiveness. Furthermore, the in-depth analysis confirms the potential of our approach to improving the effectiveness of conversational search.

cs.IR↗

arXiv2023

Line Emission Mapper (LEM): Probing the physics of cosmic ecosystems

The Line Emission Mapper (LEM) is an X-ray Probe for the 2030s that will answer the outstanding questions of the Universe's structure formation. It will also provide transformative new observing capabilities for every area of astrophysics, and to heliophysics and planetary physics as well. LEM's main goal is a comprehensive look at the physics of galaxy formation, including stellar and black-hole feedback and flows of baryonic matter into and out of galaxies. These processes are best studied in X-rays, and emission-line mapping is the pressing need in this area. LEM will use a large microcalorimeter array/IFU, covering a 30x30' field with 10" angular resolution, to map the soft X-ray line emission from objects that constitute galactic ecosystems. These include supernova remnants, star-forming regions, superbubbles, galactic outflows (such as the Fermi/eROSITA bubbles in the Milky Way and their analogs in other galaxies), the Circumgalactic Medium in the Milky Way and other galaxies, and the Intergalactic Medium at the outskirts and beyond the confines of galaxies and clusters. LEM's 1-2 eV spectral resolution in the 0.2-2 keV band will make it possible to disentangle the faintest emission lines in those objects from the bright Milky Way foreground, providing groundbreaking measurements of the physics of these plasmas, from temperatures, densities, chemical composition to gas dynamics. While LEM's main focus is on galaxy formation, it will provide transformative capability for all classes of astrophysical objects, from the Earth's magnetosphere, planets and comets to the interstellar medium and X-ray binaries in nearby galaxies, AGN, and cooling gas in galaxy clusters. In addition to pointed observations, LEM will perform a shallow all-sky survey that will dramatically expand the discovery space.

astro-ph.IM↗

arXiv2023

A Dense Representation Framework for Lexical and Semantic Matching

Lexical and semantic matching capture different successful approaches to text retrieval and the fusion of their results has proven to be more effective and robust than either alone. Prior work performs hybrid retrieval by conducting lexical and semantic matching using different systems (e.g., Lucene and Faiss, respectively) and then fusing their model outputs. In contrast, our work integrates lexical representations with dense semantic representations by densifying high-dimensional lexical representations into what we call low-dimensional dense lexical representations (DLRs). Our experiments show that DLRs can effectively approximate the original lexical representations, preserving effectiveness while improving query latency. Furthermore, we can combine dense lexical and semantic representations to generate dense hybrid representations (DHRs) that are more flexible and yield faster retrieval compared to existing hybrid techniques. In addition, we explore it jointly training lexical and semantic representations in a single model and empirically show that the resulting DHRs are able to combine the advantages of the individual components. Our best DHR model is competitive with state-of-the-art single-vector and multi-vector dense retrievers in both in-domain and zero-shot evaluation settings. Furthermore, our model is both faster and requires smaller indexes, making our dense representation framework an attractive approach to text retrieval. Our code is available at https://github.com/castorini/dhr.

cs.IR↗

arXiv2023

How to Train Your DRAGON: Diverse Augmentation Towards Generalizable Dense Retrieval

Various techniques have been developed in recent years to improve dense retrieval (DR), such as unsupervised contrastive learning and pseudo-query generation. Existing DRs, however, often suffer from effectiveness tradeoffs between supervised and zero-shot retrieval, which some argue was due to the limited model capacity. We contradict this hypothesis and show that a generalizable DR can be trained to achieve high accuracy in both supervised and zero-shot retrieval without increasing model size. In particular, we systematically examine the contrastive learning of DRs, under the framework of Data Augmentation (DA). Our study shows that common DA practices such as query augmentation with generative models and pseudo-relevance label creation using a cross-encoder, are often inefficient and sub-optimal. We hence propose a new DA approach with diverse queries and sources of supervision to progressively train a generalizable DR. As a result, DRAGON, our dense retriever trained with diverse augmentation, is the first BERT-base-sized DR to achieve state-of-the-art effectiveness in both supervised and zero-shot evaluations and even competes with models using more complex late interaction (ColBERTv2 and SPLADE++).

cs.IR↗

arXiv2022

CITADEL: Conditional Token Interaction via Dynamic Lexical Routing for Efficient and Effective Multi-Vector Retrieval

Multi-vector retrieval methods combine the merits of sparse (e.g. BM25) and dense (e.g. DPR) retrievers and have achieved state-of-the-art performance on various retrieval tasks. These methods, however, are orders of magnitude slower and need much more space to store their indices compared to their single-vector counterparts. In this paper, we unify different multi-vector retrieval models from a token routing viewpoint and propose conditional token interaction via dynamic lexical routing, namely CITADEL, for efficient and effective multi-vector retrieval. CITADEL learns to route different token vectors to the predicted lexical ``keys'' such that a query token vector only interacts with document token vectors routed to the same key. This design significantly reduces the computation cost while maintaining high accuracy. Notably, CITADEL achieves the same or slightly better performance than the previous state of the art, ColBERT-v2, on both in-domain (MS MARCO) and out-of-domain (BEIR) evaluations, while being nearly 40 times faster. Code and data are available at https://github.com/facebookresearch/dpr-scale.

cs.IR↗

arXiv2022

Strong Gravitational Lensing Parameter Estimation with Vision Transformer

Quantifying the parameters and corresponding uncertainties of hundreds of strongly lensed quasar systems holds the key to resolving one of the most important scientific questions: the Hubble constant ($H_{0}$) tension. The commonly used Markov chain Monte Carlo (MCMC) method has been too time-consuming to achieve this goal, yet recent work has shown that convolution neural networks (CNNs) can be an alternative with seven orders of magnitude improvement in speed. With 31,200 simulated strongly lensed quasar images, we explore the usage of Vision Transformer (ViT) for simulated strong gravitational lensing for the first time. We show that ViT could reach competitive results compared with CNNs, and is specifically good at some lensing parameters, including the most important mass-related parameters such as the center of lens $θ_{1}$ and $θ_{2}$, the ellipticities $e_1$ and $e_2$, and the radial power-law slope $γ'$. With this promising preliminary result, we believe the ViT (or attention-based) network architecture can be an important tool for strong lensing science for the next generation of surveys. The open source of our code and data is in \url{https://github.com/kuanweih/strong_lensing_vit_resnet}.

astro-ph.CO↗