arXiv subjects

Taejong Joo

Publications and source records attributed to Taejong Joo.

9 recordsLinked to original sources

arXiv2026

Mitigating Over-Optimization in PRM-Guided Search in Mathematical Reasoning by Optimizing the Guide

Process reward models (PRMs) provide dense step-level guidance for search-based reasoning, enabling inference-time compute to be allocated toward promising partial solutions. However, recent evidence suggests that PRM-guided search can over-optimize imperfect process rewards, pruning viable trajectories while expanding spurious ones. In this work, we theoretically show that directly leveraging PRM score is vulnerable to verifier noise through an extreme-value effect: non-viable prefixes become more likely to receive spuriously high scores as reasoning depth increase. Therefore, we formulate the PRM-guided search as a robust optimization problem over plausible reward perturbations, termed maximin PRM-guided search, leading to a training-free robust process supervision method that preserves promising alternatives when step-level scores are noisy. Maximin PRM-guided search mitigates this failure mode by reducing sensitivity to over-optimized PRM outliers. Without fine-tuning or online adaptation, maximin search consistently improves the PRM-guided search by 17-35\% on average, outperforming outcome- and step-level baselines in 14 out of 16 settings. Our source code is available at https://github.com/tjoo512/maximin-search.

cs.AI↗

arXiv2026

On Surprising Effectiveness of Masking Updates in Adaptive Optimizers

Training large language models (LLMs) relies almost exclusively on dense adaptive optimizers with increasingly sophisticated preconditioners. We challenge this by showing that randomly masking parameter updates can be highly effective, with a masked variant of RMSProp consistently outperforming recent state-of-the-art optimizers. Our analysis reveals that the random masking induces a curvature-dependent geometric regularization that smooths the optimization trajectory. Motivated by this finding, we introduce Momentum-aligned gradient masking (Magma), which modulates the masked updates using momentum-gradient alignment. Extensive LLM pre-training experiments show that Magma is a simple drop-in replacement for adaptive optimizers with consistent gains and negligible computational overhead. Notably, for the 1B model size, Magma reduces perplexity by over 19\% and 9\% compared to Adam and Muon, respectively.

cs.LG↗

arXiv2025

Graph of Agents: Principled Long Context Modeling by Emergent Multi-Agent Collaboration

As a model-agnostic approach to long context modeling, multi-agent systems can process inputs longer than a large language model's context window without retraining or architectural modifications. However, their performance often heavily relies on hand-crafted multi-agent collaboration strategies and prompt engineering, which limit generalizability. In this work, we introduce a principled framework that formalizes the model-agnostic long context modeling problem as a compression problem, yielding an information-theoretic compression objective. Building on this framework, we propose Graph of Agents (GoA), which dynamically constructs an input-dependent collaboration structure that maximizes this objective. For Llama 3.1 8B and Qwen3 8B across six document question answering benchmarks, GoA improves the average $F_1$ score of retrieval-augmented generation by 5.7\% and a strong multi-agent baseline using a fixed collaboration structure by 16.35\%, respectively. Even with only a 2K context window, GoA surpasses the 128K context window Llama 3.1 8B on LongBench, showing a dramatic increase in effective context length. Our source code is available at https://github.com/tjoo512/graph-of-agents.

cs.LG↗

arXiv2025

Technical Debt in In-Context Learning: Diminishing Efficiency in Long Context

Transformers have demonstrated remarkable in-context learning (ICL) capabilities, adapting to new tasks by simply conditioning on demonstrations without parameter updates. Compelling empirical and theoretical evidence suggests that ICL, as a general-purpose learner, could outperform task-specific models. However, it remains unclear to what extent the transformers optimally learn in-context compared to principled learning algorithms. To investigate this, we employ a meta ICL framework in which each prompt defines a distinctive regression task whose target function is drawn from a hierarchical distribution, requiring inference over both the latent model class and task-specific parameters. Within this setup, we benchmark sample complexity of ICL against principled learning algorithms, including the Bayes optimal estimator, under varying performance requirements. Our findings reveal a striking dichotomy: while ICL initially matches the efficiency of a Bayes optimal estimator, its efficiency significantly deteriorates in long context. Through an information-theoretic analysis, we show that the diminishing efficiency is inherent to ICL. These results clarify the trade-offs in adopting ICL as a universal problem solver, motivating a new generation of on-the-fly adaptive methods without the diminishing efficiency.

cs.LG↗

arXiv2024

Improving self-training under distribution shifts via anchored confidence with theoretical guarantees

Self-training often falls short under distribution shifts due to an increased discrepancy between prediction confidence and actual accuracy. This typically necessitates computationally demanding methods such as neighborhood or ensemble-based label corrections. Drawing inspiration from insights on early learning regularization, we develop a principled method to improve self-training under distribution shifts based on temporal consistency. Specifically, we build an uncertainty-aware temporal ensemble with a simple relative thresholding. Then, this ensemble smooths noisy pseudo labels to promote selective temporal consistency. We show that our temporal ensemble is asymptotically correct and our label smoothing technique can reduce the optimality gap of self-training. Our extensive experiments validate that our approach consistently improves self-training performances by 8% to 16% across diverse distribution shift scenarios without a computational overhead. Besides, our method exhibits attractive properties, such as improved calibration performance and robustness to different hyperparameter choices.

cs.LG↗

arXiv2023

IW-GAE: Importance Weighted Group Accuracy Estimation for Improved Calibration and Model Selection in Unsupervised Domain Adaptation

Distribution shifts pose significant challenges for model calibration and model selection tasks in the unsupervised domain adaptation problem -- a scenario where the goal is to perform well in a distribution shifted domain without labels. In this work, we tackle difficulties coming from distribution shifts by developing a novel importance weighted group accuracy estimator. Specifically, we present a new perspective of addressing the model calibration and model selection tasks by estimating the group accuracy. Then, we formulate an optimization problem for finding an importance weight that leads to an accurate group accuracy estimation with theoretical analyses. Our extensive experiments show that our approach improves state-of-the-art performances by 22% in the model calibration task and 14% in the model selection task.

cs.LG↗

arXiv2021

Revisiting Explicit Regularization in Neural Networks for Well-Calibrated Predictive Uncertainty

From the statistical learning perspective, complexity control via explicit regularization is a necessity for improving the generalization of over-parameterized models. However, the impressive generalization performance of neural networks with only implicit regularization may be at odds with this conventional wisdom. In this work, we revisit the importance of explicit regularization for obtaining well-calibrated predictive uncertainty. Specifically, we introduce a probabilistic measure of calibration performance, which is lower bounded by the log-likelihood. We then explore explicit regularization techniques for improving the log-likelihood on unseen samples, which provides well-calibrated predictive uncertainty. Our findings present a new direction to improve the predictive probability quality of deterministic neural networks, which can be an efficient and scalable alternative to Bayesian neural networks and ensemble methods.

cs.LG↗

arXiv2020

Being Bayesian about Categorical Probability

Neural networks utilize the softmax as a building block in classification tasks, which contains an overconfidence problem and lacks an uncertainty representation ability. As a Bayesian alternative to the softmax, we consider a random variable of a categorical probability over class labels. In this framework, the prior distribution explicitly models the presumed noise inherent in the observed label, which provides consistent gains in generalization performance in multiple challenging tasks. The proposed method inherits advantages of Bayesian approaches that achieve better uncertainty estimation and model calibration. Our method can be implemented as a plug-and-play loss function with negligible computational overhead compared to the softmax with the cross-entropy loss function.

cs.LG↗

arXiv2020

Regularizing activations in neural networks via distribution matching with the Wasserstein metric

Regularization and normalization have become indispensable components in training deep neural networks, resulting in faster training and improved generalization performance. We propose the projected error function regularization loss (PER) that encourages activations to follow the standard normal distribution. PER randomly projects activations onto one-dimensional space and computes the regularization loss in the projected space. PER is similar to the Pseudo-Huber loss in the projected space, thus taking advantage of both $L^1$ and $L^2$ regularization losses. Besides, PER can capture the interaction between hidden units by projection vector drawn from a unit sphere. By doing so, PER minimizes the upper bound of the Wasserstein distance of order one between an empirical distribution of activations and the standard normal distribution. To the best of the authors' knowledge, this is the first work to regularize activations via distribution matching in the probability distribution space. We evaluate the proposed method on the image classification task and the word-level language modeling task.

cs.LG↗