arXiv subjects

Jinyoung Park

Publications and source records attributed to Jinyoung Park.

At least 19 recordsLinked to original sources

arXiv2026

A dimension-free comparison between expectation thresholds and fractional expectation thresholds

We prove a dimension-free comparison between the expectation threshold $q(\mathcal F)$ and the fractional expectation threshold $q_f(\mathcal F)$ for any nontrivial increasing family $\mathcal F$ on a finite ground set. Specifically, we show that there is a universal constant $K>0$ such that \[ q_f(\mathcal F)\le Kq(\mathcal F)\max\{1,\log\log(1/q(\mathcal F))\}. \] Combining this comparison with a recent result of Li [arXiv:2609.08967] on the fractional version of Talagrand's discrete Convexity Conjecture, we obtain a dimension-independent bound toward the conjecture.

math.CO↗

arXiv2026

StreamAlign: Streaming Text-Aligned Speech Tokenization

Text-aligned speech tokenization methods have emerged to better align speech tokens with LLM token spaces, enabling more effective utilization of pretrained LLMs. However, they rely on offline automatic speech recognition (ASR), leading to two key limitations: (i) the need for complete utterances before tokenization, precluding real-time streaming, and (ii) vocabulary mismatch between ASR and LLMs, which reduces acoustic granularity from the subword to the word level. We introduce StreamAlign, a text-aligned speech tokenization framework that enables streaming tokenization for real-time speech-text joint modeling. StreamAlign performs online speech-text alignment by combining character-level RNN-Transducer alignment with word-level ASR guidance, mitigating ASR-LLM vocabulary mismatch while preserving recognition accuracy. A proactive word boundary classifier anticipates word completion at chunk boundaries, reducing tokenization latency from 560 ms to 270 ms. On LibriSpeech, StreamAlign achieves the lowest WER and highest UTMOS among evaluated tokenizers. Furthermore, StreamAlign-SLM, a spoken language model trained on StreamAlign units, outperforms other end-to-end spoken language models in speech continuation while achieving the strongest overall consistency on SALMon and spoken StoryCloze.

cs.CL↗

arXiv2026

Reason in the Words You Speak: Idiolectal Paraphrasing Off-Policy Traces for Reasoning Distillation in VideoLLMs

Recent large language models achieve strong performance on complex reasoning tasks, where reinforcement learning with Group Relative Policy Optimization (GRPO) has emerged as a leading paradigm for optimizing models on self-generated trajectories. However, the on-policy nature of GRPO bounds the model to the reasoning skills it can already produce, restricting to learn more advanced capabilities. Prior works inject privileged reasoning traces from a stronger teacher policy to guide training, yet these traces are inherently out of distribution with respect to the student policy. We observe that this mismatch between on-policy and off-policy causes gradient clipping on semantically critical reasoning tokens, ultimately rewarding correct answers while leaving the reasoning that justifies them unlearned. Hence, we propose \textbf{Echo-GRPO}, a framework that lets the model reason in the words it speaks. Rather than imitating low-probability privileged traces from the teacher model, Echo-GRPO rewrites them into the student policy's own \textit{idiolect}, that is, its own characteristic vocabulary and expression patterns, while preserving their semantics via Dual-Reference Decoding. We instantiate this framework as \textbf{VideoEcho-R1} for video reasoning distillation, achieving consistent improvements across three multimodal LLM backbones and five benchmarks. Finally, we show that our idiolectal paraphrasing is a plug-in module that consistently improves both RL and supervised fine-tuning frameworks for reasoning distillation, demonstrating that policy-aligned supervision extends beyond GRPO.

cs.CV↗

arXiv2026

Context Blindness in DPO: Mitigating Object Hallucination in MLLMs via Context-Calibrated Preference Optimization

Multimodal large language models (MLLMs) have made rapid progress, yet they still exhibit object hallucination, generating plausible but incorrect descriptions that are inconsistent with the visual input. Direct Preference Optimization (DPO) mitigates this by training models to prefer non-hallucinated responses over hallucinated ones, and recent efforts further enrich the preference data with relevant context. However, it remains unclear whether DPO actually leverages such context. To investigate this, we propose Contextual Preference Gain (CPG), a simple metric that measures how much a model's preference strengthens when relevant context is provided. We find that higher CPG consistently corresponds to lower hallucination, yet standard DPO and its variants exhibit only limited CPG, indicating that they underutilize contextual information and thus remain prone to hallucination. To address this, we propose Context-Calibrated DPO (C$^2$-DPO), which directly maximizes CPG while preserving the original preference ordering. Across multiple benchmarks, C$^2$-DPO substantially reduces hallucination without compromising general reasoning, relatively reducing the Object HalBench hallucination rate of Qwen2-VL-Instruct-2B by 36%. Code is available at https://github.com/mlvlab/C2-DPO

cs.CV↗

arXiv2026

A reformulation of the discrete Convexity Conjecture via $k$-thresholds

We introduce the notion of "$k$-thresholds'' and show that Talagrand's discrete convexity conjecture is equivalent to the assertion that, for some universal integer $k \ge 2$, the $k$-threshold of every increasing family is at most a universal constant times its expectation threshold. We prove a reduction theorem that bounds the $k$-threshold of any increasing graph property in terms of ordinary thresholds of graphs in suitable decompositions of its members. As a consequence, we determine, up to a constant factor, the $k$-threshold of every fixed graph in terms of a natural $k$-density parameter. We also prove that $k=2$ suffices for several classical spanning graph containment properties. More generally, we establish the conjectured comparison between $k$-thresholds and expectation thresholds for broad classes of graph containment properties whose target graphs have low degeneracy.

math.CO↗

arXiv2026

SparSEEty: Extracting Tokens from Sparsity-Exploiting LLM Serving Systems via Deterministic Side Channels

Modern large language models (LLMs) exhibit activation sparsity, wherein only a subset of their neurons is activated for given input tokens. Researchers have leveraged this property to optimize LLM serving systems by omitting weight accesses and computations pertaining to inactive neurons. Unfortunately, however, such optimizations create input-dependent weight accesses, which can be leaked over side channels. We present SparSEEty, a new token extraction attack that exploits input-dependent neuron weight accesses introduced by sparsity-exploiting LLM serving systems. SparSEEty first constructs a neuron-activation oracle using neuron weight access side channels during LLM inference, and then inverts the activation traces to reconstruct the input tokens, forming an end-to-end token extraction attack. We instantiate SparSEEty against an LLM serving system protected inside an Intel TDX confidential virtual machine (CVM), addressing three key challenges: (i) constructing a neuron-activation oracle using a combination of side channels exposed by CVMs, (ii) reducing inference-time overheads of neuron activation monitoring for covertness, and (iii) accurately inverting partial binary activation traces back to tokens. Our evaluation shows that SparSEEty can reconstruct both prompt and response tokens with consistently high BLEU scores (>0.95) across various models and datasets, while incurring monitoring overheads of 3.7% to 7.2%.

cs.CR↗

arXiv2026

HyperGraphPro: Progress-Aware Reinforcement Learning for Structure-Guided Hypergraph RAG

Graph Retrieval-Augmented Generation (GraphRAG) has emerged as a promising paradigm that organizes external knowledge into structured graphs of entities and relations, enabling large language models (LLMs) to perform complex reasoning beyond text-chunk retrieval. Recent advances have integrated reinforcement learning (RL) into agentic GraphRAG approaches, enabling iterative interactions with knowledge graphs during training. However, existing RL-based methods suffer from two key limitations: (1) they primarily depend on semantic similarity for retrieval, often overlooking the underlying graph topology, and (2) they rely on sparse, outcome-level rewards that fail to capture the quality of intermediate retrieval steps and their dependencies. To address these limitations, we propose HyperGraphPro, a progress-aware agentic framework for graph-based retrieval and multi-step reasoning. HyperGraphPro introduces a structure-aware hypergraph retrieval mechanism that jointly considers semantic relevance and graph connectivity, promoting coherent traversal along multi-hop reasoning paths. Furthermore, we design a progress-based stepwise policy optimization that provides dense learning signals by modulating advantages according to intermediate reasoning progress within a graph, rather than relying solely on final outcomes. Experiments on multi-hop question answering benchmarks demonstrate that HyperGraphPro consistently improves reasoning accuracy and generation quality over existing GraphRAG methods.

cs.CL↗

arXiv2026

Range of random $\mathbb Z$-homomorphisms on weak expanders

We prove that random $\mathbb{Z}$-homomorphisms on weakly expanding bipartite graphs exhibit a strong "flatness" phenomenon. Extending prior work of Peled, Samotij, and Yehudayoff for expanders, we first show that on any bipartite $(n, d, λ)$-graph with $λ\leq (1-δ)d$, a uniformly chosen $\mathbb{Z}$-homomorphism has a range at most $O(\log \log n)$ with high probability, which is tight up to a constant factor. This provides an affirmative answer to their question in the spectral setting. As a concrete application, we prove that a random $\mathbb{Z}$-homomorphism on the middle layers of the Hamming cube takes at most $5$ values with high probability. This shows that the $O(1)$-flatness for the full Hamming cube, proved by Kahn and Galvin, persists even when the rigid structural properties are relaxed.

math.CO↗

arXiv2026

RegFormer: Transferable Relational Grounding for Efficient Weakly-Supervised Human-Object Interaction Detection

Weakly-supervised Human-Object Interaction (HOI) detection is essential for scalable scene understanding, as it learns interactions from only image-level annotations. Due to the lack of localization signals, prior works typically rely on an external object detector to generate candidate pairs and then infer their interactions through pairwise reasoning. However, this framework often struggles to scale due to the substantial computational cost incurred by enumerating numerous instance pairs. In addition, it suffers from false positives arising from non-interactive combinations, which hinder accurate instance-level HOI reasoning. To address these issues, we introduce Relational Grounding Transformer (RegFormer), a versatile interaction recognition module for efficient and accurate HOI reasoning. Under image-level supervision, RegFormer leverages spatially grounded signals as guidance for the reasoning process and promotes locality-aware interaction learning. By learning localized interaction cues, our module distinguishes humans, objects, and their interactions, enabling direct transfer from image-level interaction reasoning to precise and efficient instance-level reasoning without additional training. Our extensive experiments and analyses demonstrate that RegFormer effectively learns spatial cues for instance-level interaction reasoning, operates with high efficiency, and even achieves performance comparable to fully supervised models. Our code is available at https://github.com/mlvlab/RegFormer.

cs.CV↗

arXiv2026

MoE-GRPO: Optimizing Mixture-of-Experts via Reinforcement Learning in Vision-Language Models

Mixture-of-Experts (MoE) has emerged as an effective approach to reduce the computational overhead of Transformer architectures by sparsely activating a subset of parameters for each token while preserving high model capacity. This paradigm has recently been extended to Vision-Language Models (VLMs), enabling scalable multi-modal understanding with reduced computational cost. However, the widely adopted deterministic top-K routing mechanism may overlook more optimal expert combinations and lead to expert overfitting. To address this limitation and improve the diversity of expert selection, we propose MoE-GRPO, a reinforcement learning (RL)-based framework for optimizing expert routing in MoE-based VLMs. Specifically, we formulate expert selection as a sequential decision-making problem and optimize it using Group Relative Policy Optimization (GRPO), allowing the model to learn adaptive expert routing policies through exploration and reward-based feedback. Furthermore, we introduce a modality-aware router guidance that enhances training stability and efficiency by discouraging the router from exploring experts that are infrequently activated for a given modality. Extensive experiments on multi-modal image and video benchmarks show that MoE-GRPO consistently outperforms standard top-K routing and its variants by promoting more diverse expert selection, thereby mitigating expert overfitting and enabling a task-level expert specialization.

cs.CV↗

arXiv2026

DeepVideo-R1: Video Reinforcement Fine-Tuning via Difficulty-aware Regressive GRPO

Recent works have demonstrated the effectiveness of reinforcement learning (RL)-based post-training for enhancing the reasoning capabilities of large language models (LLMs). In particular, Group Relative Policy Optimization (GRPO) has shown impressive success using a PPO-style reinforcement learning algorithm with group-normalized rewards. However, the effectiveness of GRPO in Video Large Language Models (VideoLLMs) remains underexplored. In this paper, we explore GRPO and identify two issues that hinder effective learning: (1) reliance on safeguards, and (2) vanishing advantage. To mitigate these challenges, we propose DeepVideo-R1, a video large language model trained with Reg-GRPO (Regressive GRPO) and difficulty-aware data augmentation. Reg-GRPO reformulates the GRPO loss function as a regression task that directly predicts the advantage in GRPO, eliminating the need for safeguards such as clipping and min operations. This directly aligns the model with the advantages, providing guidance to prefer better outputs. The difficulty-aware data augmentation strategy augments input prompts/videos to target solvable difficulty levels, enabling diverse reward signals. Our experimental results show that our approach significantly improves video reasoning performance across multiple benchmarks.

cs.CV↗

arXiv2026

Improving Large Molecular Language Model via Relation-aware Multimodal Collaboration

Large language models (LLMs) have demonstrated their instruction-following capabilities and achieved powerful performance on various tasks. Inspired by their success, recent works in the molecular domain have led to the development of large molecular language models (LMLMs) that integrate 1D molecular strings or 2D molecular graphs into the language models. However, existing LMLMs often suffer from hallucination and limited robustness, largely due to inadequate integration of diverse molecular modalities such as 1D sequences, 2D molecular graphs, and 3D conformations. To address these limitations, we propose CoLLaMo, a large language model-based molecular assistant equipped with a multi-level molecular modality-collaborative projector. The relation-aware modality-collaborative attention mechanism in the projector facilitates fine-grained and relation-guided information exchange between atoms by incorporating 2D structural and 3D spatial relations. Furthermore, we present a molecule-centric new automatic measurement, including a hallucination assessment metric and GPT-based caption quality evaluation to address the limitations of token-based generic evaluation metrics (i.e., BLEU) widely used in assessing molecular comprehension of LMLMs. Our extensive experiments demonstrate that our CoLLaMo enhances the molecular modality generalization capabilities of LMLMs, achieving the best performance on multiple tasks, including molecule captioning, computed property QA, descriptive property QA, motif counting, and IUPAC name prediction.

cs.AI↗

arXiv2026

A refined graph container lemma and applications to the hard-core model on bipartite expanders

We establish a refined version of a graph container lemma due to Galvin and discuss several applications related to the hard-core model on bipartite expander graphs. Given a graph $G$ and $λ>0$, the hard-core model on $G$ at activity $λ$ is the probability distribution $μ_{G,λ}$ on independent sets in $G$ given by $μ_{G,λ}(I)\propto λ^{|I|}$. As one of our main applications, we show that the hard-core model at activity $λ$ on the hypercube $Q_d$ exhibits a `structured phase' for $λ= Ω( \log^2 d/d^{1/2})$ in the following sense: in a typical sample from $μ_{Q_d,λ}$, most vertices are contained in one side of the bipartition of $Q_d$. This improves upon a result of Galvin which establishes the same for $λ=Ω(\log d/ d^{1/3})$. As another application, we establish a fully polynomial-time approximation scheme (FPTAS) for the hard-core model on a $d$-regular bipartite $α$-expander, with $α>0$ fixed, when $λ= Ω( \log^2 d/d^{1/2})$. This improves upon the bound $λ=Ω(\log d/ d^{1/4})$ due to the first author, Perkins and Potukuchi. We discuss similar improvements to results of Galvin-Tetali, Balogh-Garcia-Li and Kronenberg-Spinka.

math.CO↗

arXiv2026

On the number of antichains in $\{0,1,2\}^n$

We provide precise asymptotics for the number of antichains in the poset $\{0,1,2\}^n$, answering a question of Sapozhenko. Finding improved estimates for this number was also a problem suggested by Noel, Scott, and Sudakov, who obtained asymptotics for the logarithm of the number. Key ingredients for the proof include a graph-container lemma to bound the number of expanding sets in a class of irregular graphs, isoperimetric inequalities for generalizations of the Boolean lattice, and methods from statistical physics based on the cluster expansion.

math.CO↗

arXiv2025

LG Uplus System with Multi-Speaker IDs and Discriminator-based Sub-Judges for the WildSpoof Challenge

This paper describes our submission to the WildSpoof Challenge Track 2, which focuses on spoof-aware speaker verification (SASV) in the presence of high-quality text-to-speech (TTS) attacks. We adopt a ResNet-221 back-bone and study two speaker-labeling strategies, namelyDual-Speaker IDs and Multi-Speaker IDs, to explicitly enlarge the margin between bona fide and generated speech in the embedding space. In addition, we propose discriminator-based sub-judge systems that reuse internal features from HiFi-GAN and BigVGAN discriminators, aggregated via multi-query multi-head attentive statistics pooling(MQMHA). Experimental results on the SpoofCeleb corpus show that our system design is effective in improving agnostic detection cost function (a-DCF).

eess.AS↗

arXiv2025

Asymptotic enumeration via graph containers and entropy

The container methods are powerful tools to bound the number of independent sets of graphs and hypergraphs, and they have been extremely influential in the area of extremal and probabilistic combinatorics. We will focus on more specialized graph container methods due to Sapozhenko (1987) that deal with sets in expander graphs. Entropy, first introduced by Shannon (1948) in the area of information theory, is a measure of the expected amount of information contained in a random variable. Entropy has seen lots of fascinating applications in a wide range of enumeration problems. In this survey article, we will discuss recent developments that exploit a combination of the two methods on enumerating graph homomorphisms.

math.CO↗

arXiv2025

On the "second" Kahn--Kalai Conjecture: cliques, cycles, and trees

We prove a few simple cases of a random graph statement that would imply the "second" Kahn--Kalai Conjecture. Even these cases turn out to be reasonably challenging, and it is hoped that the ideas introduced here may lead to further interest in, and further progress on, this natural problem.

math.CO↗

arXiv2025

Towards Efficient Vision State Space Models via Token Merging

State Space Models (SSMs) have emerged as powerful architectures in computer vision, yet improving their computational efficiency remains crucial for practical and scalable deployment.While token reduction serves as an effective approach for model efficiency, applying it to SSMs requires careful consideration of their unique sequential modeling capabilities.In this work, we propose MaMe, a token-merging strategy tailored for SSM-based vision models.MaMe addresses two key challenges: quantifying token importance and preserving sequential properties. Our approach leverages the state transition parameter $\mathbfΔ$ as an informativeness measure and introduces strategic token arrangements to preserve sequential information flow.Extensive experiments demonstrate that MaMe achieves superior efficiency-performance trade-offs for both fine-tuned and off-the-shelf models. Particularly, our approach maintains robustness even under aggressive token reduction where existing methods undergo significant performance degradation.Beyond image classification, MaMe shows strong generalization capabilities across video and audio domains, establishing an effective approach for enhancing efficiency in diverse SSM applications.

cs.CV↗