arXiv subjects

Allen Lin

Publications and source records attributed to Allen Lin.

13 recordsLinked to original sources

arXiv2026

Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA

Macaron-V1 is an open agent-model family for experiential intelligence: learning from experience in real environments and continuing to learn after deployment. It is organized around two system goals. Adaptation is pursued through recursive improvement of versioned model-harness pairs, where experience from one configuration is evaluated under an external contract and used to construct its successor. Collaboration is pursued via the Mixture-of-LoRA (MoL) architecture that freezes a base model, composes specialist LoRA adapters, and selects one LoRA per user turn. The flagship Macaron-V1-Venti (748B) combines a 744B GLM-5.2 base with four LoRAs for chat, agent, coding, and GenUI; the Qwen3.6-35B-based Macaron-V1-Tall (50B) uses the same design for local deployment. This report presents Macaron-V1 as a co-designed system spanning architecture, algorithms, and infrastructure. The MoL architecture supports continual learning through extensible LoRA specialists. The algorithm combines Model-Harness Co-design and recursive self-improvement loop, including the UI4A component-native GenUI harness, a stateful action substrate, versioned Harness Context Protocol contract, and the agentic RL framework MindForge. The supporting infrastructure includes the post-training platform MinT, the long-context RL method LongStraw, and stability techniques for sparse MoE and DSA base models. We evaluate Macaron-V1 on Personal Intelligence, GenUI, and general capability benchmarks against frontier baselines. Our results validate the current system, while compounding gains from continual learning and collective intelligence remain open questions.

cs.LG↗

arXiv2026

On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters

Parameter-efficient fine-tuning (PEFT) is usually treated as a cheaper alternative to full fine-tuning. We study a broader role: small trainable adapters as persistent local state on top of strong shared foundation models. In this framing, the base model provides shared competence while adapters carry instance-specific behavior such as preferences, skills, tool habits, and memory-like updates. We organize the problem around three scaling axes: Scale Up, where stronger shared priors make small local updates more useful; Scale Down, where we study how small adapters can be while remaining reliable; and Scale Out, where many persistent adapted instances coexist. MinT provides one infrastructure example for managing adapter identity, revision, provenance, evaluation, and serving residency. Together, the results suggest that PEFT can be a compact substrate for persistent personal models rather than only a budget substitute for full fine-tuning.

cs.LG↗

arXiv2025

DMRetriever: A Family of Models for Improved Text Retrieval in Disaster Management

Effective and efficient access to relevant information is essential for disaster management. However, no retrieval model is specialized for disaster management, and existing general-domain models fail to handle the varied search intents inherent to disaster management scenarios, resulting in inconsistent and unreliable performance. To this end, we introduce DMRetriever, the first series of dense retrieval models (33M to 7.6B) tailored for this domain. It is trained through a novel three-stage framework of bidirectional attention adaptation, unsupervised contrastive pre-training, and difficulty-aware progressive instruction fine-tuning, using high-quality data generated through an advanced data refinement pipeline. Comprehensive experiments demonstrate that DMRetriever achieves state-of-the-art (SOTA) performance across all six search intents at every model scale. Moreover, DMRetriever is highly parameter-efficient, with 596M model outperforming baselines over 13.3 X larger and 33M model exceeding baselines with only 7.6% of their parameters. All codes, data, and checkpoints are available at https://github.com/KaiYin97/DMRETRIEVER

cs.IR↗

arXiv2025

Optimal Thresholds for Monotone Non-Boolean Functions

Let $[q] = \{0,1,\ldots,q-1\}$, let $\Delta[q]$ denote the simplex of probability measures on $[q]$, and let $\gamma$ denote the Lebesgue measure normalized on $\Delta[q]$. We prove that for any symmetric monotone function $f \colon[q]^n \to [q]$ and any $a \in [q]$ we have \begin{equation*} \gamma(\{\mu \in \Delta[q]\;\vert\;\mathbb{P}_{x\sim\mu^{\otimes n}}[f(x)=a] \in (\varepsilon,1-\varepsilon)\}) = O(1/\log n)\text{.} \end{equation*} We also show that this bound is tight. This improves Kalai and Mossel's previous bound of $O(\log \log n/\log n)$ and answers their question completely.

math.PR↗

arXiv2025

Federated Conversational Recommender System

Conversational Recommender Systems (CRSs) have become increasingly popular as a powerful tool for providing personalized recommendation experiences. By directly engaging with users in a conversational manner to learn their current and fine-grained preferences, a CRS can quickly derive recommendations that are relevant and justifiable. However, existing conversational recommendation systems (CRSs) typically rely on a centralized training and deployment process, which involves collecting and storing explicitly-communicated user preferences in a centralized repository. These fine-grained user preferences are completely human-interpretable and can easily be used to infer sensitive information (e.g., financial status, political stands, and health information) about the user, if leaked or breached. To address the user privacy concerns in CRS, we first define a set of privacy protection guidelines for preserving user privacy under the conversational recommendation setting. Based on these guidelines, we propose a novel federated conversational recommendation framework that effectively reduces the risk of exposing user privacy by (i) de-centralizing both the historical interests estimation stage and the interactive preference elicitation stage and (ii) strictly bounding privacy leakage by enforcing user-level differential privacy with meticulously selected privacy budgets. Through extensive experiments, we show that the proposed framework not only satisfies these user privacy protection guidelines, but also enables the system to achieve competitive recommendation performance even when compared to the state-of-the-art non-private conversational recommendation approach.

cs.IR↗

arXiv2025

Towards An Efficient LLM Training Paradigm for CTR Prediction

Large Language Models (LLMs) have demonstrated tremendous potential as the next-generation ranking-based recommendation system. Many recent works have shown that LLMs can significantly outperform conventional click-through-rate (CTR) prediction approaches. Despite such promising results, the computational inefficiency inherent in the current training paradigm makes it particularly challenging to train LLMs for ranking-based recommendation tasks on large datasets. To train LLMs for CTR prediction, most existing studies adopt the prevalent ''sliding-window'' paradigm. Given a sequence of $m$ user interactions, a unique training prompt is constructed for each interaction by designating it as the prediction target along with its preceding $n$ interactions serving as context. In turn, the sliding-window paradigm results in an overall complexity of $O(mn^2)$ that scales linearly with the length of user interactions. Consequently, a direct adoption to train LLMs with such strategy can result in prohibitively high training costs as the length of interactions grows. To alleviate the computational inefficiency, we propose a novel training paradigm, namely Dynamic Target Isolation (DTI), that structurally parallelizes the training of $k$ (where $k >> 1$) target interactions. Furthermore, we identify two major bottlenecks - hidden-state leakage and positional bias overfitting - that limit DTI to only scale up to a small value of $k$ (e.g., 5) then propose a computationally light solution to effectively tackle each. Through extensive experiments on three widely adopted public CTR datasets, we empirically show that DTI reduces training time by an average of $\textbf{92%}$ (e.g., from $70.5$ hrs to $5.31$ hrs), without compromising CTR prediction performance.

cs.IR↗

arXiv2025

Cold-Start Recommendation towards the Era of Large Language Models (LLMs): A Comprehensive Survey and Roadmap

Cold-start problem is one of the long-standing challenges in recommender systems, focusing on accurately modeling new or interaction-limited users or items to provide better recommendations. Due to the diversification of internet platforms and the exponential growth of users and items, the importance of cold-start recommendation (CSR) is becoming increasingly evident. At the same time, large language models (LLMs) have achieved tremendous success and possess strong capabilities in modeling user and item information, providing new potential for cold-start recommendations. However, the research community on CSR still lacks a comprehensive review and reflection in this field. Based on this, in this paper, we stand in the context of the era of large language models and provide a comprehensive review and discussion on the roadmap, related literature, and future directions of CSR. Specifically, we have conducted an exploration of the development path of how existing CSR utilizes information, from content features, graph relations, and domain information, to the world knowledge possessed by large language models, aiming to provide new insights for both the research and industrial communities on CSR. Related resources of cold-start recommendations are collected and continuously updated for the community in https://github.com/YuanchenBei/Awesome-Cold-Start-Recommendation.

cs.IR↗

arXiv2024

The Gauss Circle Problem and Fourier Quasicrystals

The Gauss circle problem asks for an approximation to the number of lattice points of $\mathbb{Z}^2$ contained in $B_r$, the disk of radius $r$ centered at the origin. Upper, lower, and average bounds have been established for this number-theoretic problem and have been generalized to any lattice in any dimension. We extend this problem to a more general class of structures known as Fourier quasicrystals. Recent work from Alon, Kummer, Kurasov, and Vinzant provides an upper bound $\#(\Lambda \cap B_r) = c_0\textrm{Vol}_d\left(B_r\right)+ O\left(r^{d-1}\right)$ for any Fourier quasicrystal $\Lambda \subset \mathbb{R}^d$ of density $c_0$, where $B_r$ is the $d$-dimensional ball of radius $r$. In this paper, we improve the upper bound for any uniformly discrete Fourier quasicrystal, by showing we can write $\#\left(\Lambda \cap B_r\right) = c_0\textrm{Vol}_d\left(B_r\right) + O\left(r^{\theta(\Lambda)}\right)$, where $\frac{d-1}{2} < \theta(\Lambda) < d-1$ is some exponent depending on $\Lambda$. In the special case $d = 2$, we also prove lower and upper bounds for the average of the error.

math-ph↗

arXiv2024

Countering Mainstream Bias via End-to-End Adaptive Local Learning

Collaborative filtering (CF) based recommendations suffer from mainstream bias -- where mainstream users are favored over niche users, leading to poor recommendation quality for many long-tail users. In this paper, we identify two root causes of this mainstream bias: (i) discrepancy modeling, whereby CF algorithms focus on modeling mainstream users while neglecting niche users with unique preferences; and (ii) unsynchronized learning, where niche users require more training epochs than mainstream users to reach peak performance. Targeting these causes, we propose a novel end-To-end Adaptive Local Learning (TALL) framework to provide high-quality recommendations to both mainstream and niche users. TALL uses a loss-driven Mixture-of-Experts module to adaptively ensemble experts to provide customized local models for different users. Further, it contains an adaptive weight module to synchronize the learning paces of different users by dynamically adjusting weights in the loss. Extensive experiments demonstrate the state-of-the-art performance of the proposed model. Code and data are provided at \url{https://github.com/JP-25/end-To-end-Adaptive-Local-Leanring-TALL-}

cs.IR↗

arXiv2023

NTIRE 2023 Quality Assessment of Video Enhancement Challenge

This paper reports on the NTIRE 2023 Quality Assessment of Video Enhancement Challenge, which will be held in conjunction with the New Trends in Image Restoration and Enhancement Workshop (NTIRE) at CVPR 2023. This challenge is to address a major challenge in the field of video processing, namely, video quality assessment (VQA) for enhanced videos. The challenge uses the VQA Dataset for Perceptual Video Enhancement (VDPVE), which has a total of 1211 enhanced videos, including 600 videos with color, brightness, and contrast enhancements, 310 videos with deblurring, and 301 deshaked videos. The challenge has a total of 167 registered participants. 61 participating teams submitted their prediction results during the development phase, with a total of 3168 submissions. A total of 176 submissions were submitted by 37 participating teams during the final testing phase. Finally, 19 participating teams submitted their models and fact sheets, and detailed the methods they used. Some methods have achieved better results than baseline methods, and the winning methods have demonstrated superior prediction performance.

cs.CV↗

arXiv2023

Enhancing User Personalization in Conversational Recommenders

Conversational recommenders are emerging as a powerful tool to personalize a user's recommendation experience. Through a back-and-forth dialogue, users can quickly hone in on just the right items. Many approaches to conversational recommendation, however, only partially explore the user preference space and make limiting assumptions about how user feedback can be best incorporated, resulting in long dialogues and poor recommendation performance. In this paper, we propose a novel conversational recommendation framework with two unique features: (i) a greedy NDCG attribute selector, to enhance user personalization in the interactive preference elicitation process by prioritizing attributes that most effectively represent the actual preference space of the user; and (ii) a user representation refiner, to effectively fuse together the user preferences collected from the interactive elicitation process to obtain a more personalized understanding of the user. Through extensive experiments on four frequently used datasets, we find the proposed framework not only outperforms all the state-of-the-art conversational recommenders (in terms of both recommendation performance and conversation efficiency), but also provides a more personalized experience for the user under the proposed multi-groundtruth multi-round conversational recommendation setting.

cs.IR↗

arXiv2022

Towards Fair Conversational Recommender Systems

Conversational recommender systems have demonstrated great success. They can accurately capture a user's current detailed preference -- through a multi-round interaction cycle -- to effectively guide users to a more personalized recommendation. Alas, conversational recommender systems can be plagued by the adverse effects of bias, much like traditional recommenders. In this work, we argue for increased attention on the presence of and methods for counteracting bias in these emerging systems. As a starting point, we propose three fundamental questions that should be deeply examined to enable fairness in conversational recommender systems.

cs.IR↗

arXiv2022

Quantifying and Mitigating Popularity Bias in Conversational Recommender Systems

Conversational recommender systems (CRS) have shown great success in accurately capturing a user's current and detailed preference through the multi-round interaction cycle while effectively guiding users to a more personalized recommendation. Perhaps surprisingly, conversational recommender systems can be plagued by popularity bias, much like traditional recommender systems. In this paper, we systematically study the problem of popularity bias in CRSs. We demonstrate the existence of popularity bias in existing state-of-the-art CRSs from an exposure rate, a success rate, and a conversational utility perspective, and propose a suite of popularity bias metrics designed specifically for the CRS setting. We then introduce a debiasing framework with three unique features: (i) Popularity-Aware Focused Learning to reduce the popularity-distorting impact on preference prediction; (ii) Cold-Start Item Embedding Reconstruction via Attribute Mapping, to improve the modeling of cold-start items; and (iii) Dual-Policy Learning, to better guide the CRS when dealing with either popular or unpopular items. Through extensive experiments on two frequently used CRS datasets, we find the proposed model-agnostic debiasing framework not only mitigates the popularity bias in state-of-the-art CRSs but also improves the overall recommendation performance.

cs.IR↗