arXiv subjects

Jiachen Yuan

Publications and source records attributed to Jiachen Yuan.

11 recordsLinked to original sources

arXiv2026

Critical embeddings

Hayut and the first author isolated the notion of a critical cardinal in [1]. In this work, we answer several questions raised in the original paper. We show that it is consistent for a critical cardinal not to have any ultrapower elementary embeddings, as well as that it is consistent that no target model is closed. We also prove that if $κ$ is a critical point by any ultrapower embedding, then it is the critical point of an ultrapower embedding by a normal measure. The paper concludes by presenting several open questions of interest in the study of critical cardinals.

math.LO↗

arXiv2026

VTC: DNN Compilation with Virtual Tensors for Data Movement Elimination

With the widening gap between compute and memory operation latencies, data movement optimizations have become increasingly important for DNN compilation. Current optimizations such as layout transformations and operator fusion only target a subset of tensor operators and consequently miss important opportunities for reducing data movement in contemporary DNN workloads, including large language models. We introduce VTC, a novel tensor compilation framework that for the first time eliminates all unnecessary data movement by targeting the full spectrum of data movement operators. VTC proposes the concept of virtual tensors to track data movement between compute operators via index mappings rather than expensive physical data transfers to and from global memory, which can seamlessly interoperate with existing computation kernels and handle arbitrary tensor operator compositions. We also introduce a novel data movement elimination algorithm to automatically identify a profitable virtual tensor creation strategy. Evaluation on a variety of DNNs shows that VTC can outperform existing ML compilers by up to 1.93x (1.28x on average) on NVIDIA GPUs with up to 60% (17.5% on average) inference memory savings.

cs.DC↗

arXiv2026

D-Mem: A Dual-Process Memory System for LLM Agents

Driven by the development of persistent, self-adapting autonomous agents, equipping these systems with high-fidelity memory access for long-horizon reasoning has emerged as a critical requirement. However, prevalent retrieval-based memory frameworks often follow an incremental processing paradigm that continuously extracts and updates conversational memories into vector databases, relying on semantic retrieval when queried. While this approach is fast, it inherently relies on lossy abstraction, frequently missing contextually critical information and struggling to resolve queries that rely on fine-grained contextual understanding. To address this, we introduce D-Mem, a dual-process memory system. It retains lightweight vector retrieval for routine queries while establishing an exhaustive Full Deliberation module as a high-fidelity fallback. To achieve cognitive economy without sacrificing accuracy, D-Mem employs a Multi-dimensional Quality Gating policy to dynamically bridge these two processes. Experiments on the LoCoMo and RealTalk benchmarks using GPT-4o-mini and Qwen3-235B-Instruct demonstrate the efficacy of our approach. Notably, our Multi-dimensional Quality Gating policy achieves an F1 score of 53.5 on LoCoMo with GPT-4o-mini. This outperforms our static retrieval baseline, Mem0$^\ast$ (51.2), and recovers 96.7\% of the Full Deliberation's performance (55.3), while incurring significantly lower computational costs.

cs.AI↗

arXiv2026

HMS-BERT: Hybrid Multi-Task Self-Training for Multilingual and Multi-Label Cyberbullying Detection

Cyberbullying on social media is inherently multilingual and multi-faceted, where abusive behaviors often overlap across multiple categories. Existing methods are commonly limited by monolingual assumptions or single-task formulations, which restrict their effectiveness in realistic multilingual and multi-label scenarios. In this paper, we propose HMS-BERT, a hybrid multi-task self-training framework for multilingual and multi-label cyberbullying detection. Built upon a pretrained multilingual BERT backbone, HMS-BERT integrates contextual representations with handcrafted linguistic features and jointly optimizes a fine-grained multi-label abuse classification task and a three-class main classification task. To address labeled data scarcity in low-resource languages, an iterative self-training strategy with confidence-based pseudo-labeling is introduced to facilitate cross-lingual knowledge transfer. Experiments on four public datasets demonstrate that HMS-BERT achieves strong performance, attaining a macro F1-score of up to 0.9847 on the multi-label task and an accuracy of 0.6775 on the main classification task. Ablation studies further verify the effectiveness of the proposed components.

cs.CL↗

arXiv2025

DyFuLM: An Advanced Multimodal Framework for Sentiment Analysis

Understanding sentiment in complex textual expressions remains a fundamental challenge in affective computing. To address this, we propose a Dynamic Fusion Learning Model (DyFuLM), a multimodal framework designed to capture both hierarchical semantic representations and fine-grained emotional nuances. DyFuLM introduces two key moodules: a Hierarchical Dynamic Fusion module that adaptively integrates multi-level features, and a Gated Feature Aggregation module that regulates cross-layer information ffow to achieve balanced representation learning. Comprehensive experiments on multi-task sentiment datasets demonstrate that DyFuLM achieves 82.64% coarse-grained and 68.48% fine-grained accuracy, yielding the lowest regression errors (MAE = 0.0674, MSE = 0.0082) and the highest R^2 coefficient of determination (R^2= 0.6903). Furthermore, the ablation study validates the effectiveness of each module in DyFuLM. When all modules are removed, the accuracy drops by 0.91% for coarse-grained and 0.68% for fine-grained tasks. Keeping only the gated fusion module causes decreases of 0.75% and 0.55%, while removing the dynamic loss mechanism results in drops of 0.78% and 0.26% for coarse-grained and fine-grained sentiment classification, respectively. These results demonstrate that each module contributes significantly to feature interaction and task balance. Overall, the experimental findings further validate that DyFuLM enhances sentiment representation and overall performance through effective hierarchical feature fusion.

cs.CL↗

arXiv2025

Ketonen's question and other cardinal sins

Answering a question of Ketonen from the late 1970's, it is proved that a weakly compact cardinal carrying an indecomposable ultrafilter need not be measurable. The result is obtained by analyzing the limit of a decreasing sequence of models of ZFC. The utility of this proof technique is demonstrated further in this paper, where a problem by Bagaria and Magidor concerning strong compactness, and a problem by Lambie-Hanson and Rinot concerning the $C$-sequence number are solved as well.

math.LO↗

arXiv2025

A new model for all $C$-sequences are trivial

We construct a model in which all $C$-sequences are trivial, yet there exists a $κ$-Souslin tree with full vanishing levels. This answers a question of Lambie-Hanson and Rinot, and provides an optimal combination of compactness and incompactness. It is obtained by incorporating a so-called mutually exclusive ascent path to Kunen's original forcing construction.

math.LO↗

arXiv2023

How far is almost strong compactness from strong compactness

Bagaria and Magidor introduced the notion of almost strong compactness, which is very close to the notion of strong compactness. Boney and Brooke-Taylor asked whether the least almost strongly compact cardinal is strongly compact. Goldberg gives a positive answer in the case $\mathrm{SCH}$ holds from below and the least almost strongly compact cardinal has uncountable cofinality. In this paper, we give a negative answer for the general case. Our result also gives an affirmative answer to a question of Bagaria and Magidor.

math.LO↗

arXiv2022

FogROS G: Enabling Secure, Connected and Mobile Fog Robotics with Global Addressability

Fog Robotics renders networked robots with greater mobility, on-demand compute capabilities and better energy efficiency by offloading heavy robotics workloads to nearby Edge and distant Cloud data centers. However, as the de-facto standard for implementing fog robotics applications, Robot Operating System (ROS) and its successor ROS2 fail to provide fog robots with a mobile-friendly and secure communication infrastructure. In this work, we present FogROS G, a secure routing framework that connects robotics software components from different physical locations, networks, Data Distribution Service (DDS) and ROS distributions. FogROS G indexes networked robots with globally unique 256-bit names that remains constant even if the robot roams between multiple administrative network domains. FogROS G leverages Global Data Plane, a global and secure peer-to-peer routing infrastructure between the names, guaranteeing that only authenticated party can send to or receive from the robot. FogROS G adopts a proxy-based design that connect nodes from ROS1 and ROS2 with mainstream DDS vendors; this can be done without any changes to the application code.

cs.RO↗

arXiv2022

On the cofinality of the least $λ$-strongly compact cardinal

In this paper, we characterize the possible cofinalities of the least $λ$-strongly compact cardinal. We show that, on the one hand, for any regular cardinal, $δ$, that carries a $λ$-complete uniform ultrafilter, it is consistent, relative to the existence of a supercompact cardinal above $δ$, that the least $λ$-strongly compact cardinal has cofinality $δ$. On the other hand, provably the cofinality of the least $λ$-strongly compact cardinal always carries a $λ$-complete uniform ultrafilter.

math.LO↗

arXiv2018

Factorials of infinite cardinals in ZF

For a set $x$, let $\mathcal{S}(x)$ be the set of all permutations of $x$. We study several aspects of this notion in $\mathsf{ZF}$. The main results are as follows: (1) $\mathsf{ZF}$ proves that for all sets $x$, if $\mathcal{S}(x)$ is Dedekind infinite, then there are no finite-to-one maps from $\mathcal{S}(x)$ into $\mathcal{S}_{\mathrm{fin}}(x)$, where $\mathcal{S}_{\mathrm{fin}}(x)$ is the set of all permutations of $x$ which move only finitely many elements. (2) $\mathsf{ZF}$ proves that for all sets $x$, the cardinality of $\mathcal{S}(x)$ is strictly greater than that of $[x]^2$. (3) It is consistent with $\mathsf{ZF}$ that there exists an infinite set $x$ such that the cardinality of $\mathcal{S}(x)$ is strictly less than that of $[x]^3$. (4) It is consistent with $\mathsf{ZF}$ that there exists an infinite set $x$ such that there is a finite-to-one map from $\mathcal{S}(x)$ into $x$.

math.LO↗