Searcharxiv⌕ Search

arXiv subjects

Changhee Joo

Publications and source records attributed to Changhee Joo.

12 recordsLinked to original sources

Retain-Neutral Surrogates for Min-Max Unlearning

Machine unlearning aims to make a model forget designated training data while retaining performance on the rest. In min-max unlearning, an inner adversary perturbs the model to maximize the forget loss, and the model is updated to minimize the retain loss at the resulting surrogate point. When the forget and retain gradients are positively coupled, the standard perturbation also moves toward higher retain loss, and the retain step begins by recovering that loss. We propose Retain-Orthogonal Surrogate Unlearning (ROSU), which redefines the adversary to admit only perturbations orthogonal to the retain gradient. The one that raises the forget loss fastest has a closed form, and ROSU uses it to set the surrogate point and amplifies it into the forgetting step. Under positive coupling the ROSU surrogate avoids the retain cost that coupling induces in the standard surrogate, and the difference shrinks as the gradients approach orthogonality. ROSU has the smallest accuracy gap to retraining in every vision setting (CIFAR-10, CIFAR-100, Tiny-ImageNet) and is competitive with the baselines on language benchmarks (TOFU, WMDP). The redefinition helps most under high positive coupling. Code is available at https://github.com/unlearn-ml/machine-unlearning.

cs.LG↗

Data-Efficient Unlearning-Based Backdoor Detection in Federated Learning

Detecting backdoor attacks in federated learning (FL) is challenging because malicious clients can manipulate the global model while preserving main-task performance. Static model-update statistics may therefore fail to clearly distinguish malicious from benign clients. Unlearning-based detectors instead probe client models through loss ascent on clean server data, but detection performance degrades with scarce proxy data. We propose Screening via Unlearning Response Analysis (SURA), a data-efficient detector. SURA reuses the same small clean proxy across multiple train-mode loss-ascent steps, strengthening client responses without additional clean data. For federated vision models with batch normalization (BN), repeated probing also changes BN running statistics and can confound endpoint measurements. SURA addresses this by restoring the shared pre-probe BN state only at readout while preserving the parameter changes induced by probing. This separates repeated probing from response measurement and removes BN drift and the interaction between parameter and BN-state changes from the readout. Across six backdoor attacks, SURA uses only 10 clean server images, outperforms methods under the same data budget, and approaches the robustness obtained with much larger proxy sets.

cs.CR↗

Spectral Unlearning: Transformer Structure-Preserving Updates for Language Models

Machine unlearning in large language models (LLMs) aims to remove specific learned knowledge, capabilities, or behaviors while preserving model utility. Existing methods have mostly focused on the design of unlearning objectives, while largely overlooking the intrinsic matrix structure of transformer weights. In particular, standard optimizers operate on independent scalar coordinates, thereby ignoring this matrix structure. We propose spectral unlearning, an update framework that combines standard unlearning objectives with matrix-structured optimization. We show that the spectral update corresponds to the steepest-descent direction for transformers composed of linear layers with normalized inputs and outputs. Our approach is applied on top of existing unlearning methods and evaluated across complementary axes of memorization, privacy, and utility, summarized via their harmonic mean. Spectral unlearning consistently outperforms a broad range of baselines and closes most of the gap to a retain-trained reference. Code is available at https://github.com/unlearn-ml/spectral-unlearning/.

cs.LG↗

On the Depth Scalability of Logic Gate Networks

Logic Gate Networks (LGNs) compute through compositions of Boolean operations, yet existing LGNs do not reliably benefit from increased depth. We identify two causes: optimization collapse and topology-induced degradation of output-specific credit that persists even after skip-biased initialization and straight-through estimation stabilize training. We introduce Input-Anchored Logic Gate Networks (IALGNs), in which each gate combines a private hidden spine with a direct input anchor. This topology prevents output-path merging while retaining input access at every layer. Credit diagnostics show that random wiring dilutes or conflicts output-specific gradients, whereas IALGN maintains usable and coherent credit. Random-$k_x$ relaxation improves anchor selection without relaxing the spine. Across MNIST, CIFAR-10, and CIFAR-100, IALGN exhibits consistent fixed-width depth--accuracy scaling up to 150 layers, while alternative topologies saturate or degrade. Linear probes, topology ablations, and operation-aware analysis show that trained IALGNs preserve private states and apply sparse anchor-conditioned updates. These results indicate that scalable LGN depth requires both stable optimization and credit-preserving information access.

cs.LG↗

SCOPE: Entanglement Frontier Escape for Source-Free Class Unlearning

Source-free class unlearning erases whole classes using only the forget data, judged at the representation level, where features can leak a class the head no longer predicts. Existing feature-space erasers answer with one fixed projection, yet forget and retain classes share a representation, so deleting one disturbs the other where they overlap. We prove this tension is a frontier. Every fixed projection that deletes pays a retain cost of at least the retain-readout energy along the forget-discriminant subspace, and erasing that subspace alone attains the floor. The leading source-free erasers all instantiate the form it binds, so the frontier limits the whole class. Conditioning the erasure on the input escapes it. Spectral Conditional Projective Erasure (SCOPE) does so with a single gate, suppressing the forget subspace chiefly on inputs its frozen head's weight scores read as a forget class. It is closed form, needs no retain data or gradient training, and costs orders of magnitude less than retraining. Across five object, face, and speaker benchmarks spanning two modalities and both convolutional and transformer backbones, the frontier predicts the measured retain cost. SCOPE leads the source-free erasers on every benchmark and forget-set size, and at the hardest setting it tops every unlearner, trained methods included.

cs.LG↗

HCF: Hierarchical Cascade Framework for Distributed Multi-Stage Image Compression

Distributed multi-stage image compression -- where visual content traverses multiple processing nodes under varying quality requirements -- poses challenges. Progressive methods enable bitstream truncation but underutilize available compute resources; successive compression repeats costly pixel-domain operations and suffers cumulative quality loss and inefficiency; fixed-parameter models lack post-encoding flexibility. In this work, we developed the Hierarchical Cascade Framework (HCF) that achieves high rate-distortion performance and better computational efficiency through direct latent-space transformations across network nodes in distributed multi-stage image compression systems. Under HCF, we introduced policy-driven quantization control to optimize rate-distortion trade-offs, and established the edge quantization principle through differential entropy analysis. The configuration based on this principle demonstrates up to 0.6dB PSNR gains over other configurations. When comprehensively evaluated on the Kodak, CLIC, and CLIC2020-mobile datasets, HCF outperforms successive-compression methods by up to 5.56% BD-Rate in PSNR on CLIC, while saving up to 97.8% FLOPs, 96.5% GPU memory, and 90.0% execution time. It also outperforms state-of-the-art progressive compression methods by up to 12.64% BD-Rate on Kodak and enables retraining-free cross-quality adaptation with 7.13-10.87% BD-Rate reductions on CLIC2020-mobile.

cs.CV↗

Geographical Node Clustering and Grouping to Guarantee Data IIDness in Federated Learning

Federated learning (FL) is a decentralized AI mechanism suitable for a large number of devices like in smart IoT. A major challenge of FL is the non-IID dataset problem, originating from the heterogeneous data collected by FL participants, leading to performance deterioration of the trained global model. There have been various attempts to rectify non-IID dataset, mostly focusing on manipulating the collected data. This paper, however, proposes a novel approach to ensure data IIDness by properly clustering and grouping mobile IoT nodes exploiting their geographical characteristics, so that each FL group can achieve IID dataset. We first provide an experimental evidence for the independence and identicalness features of IoT data according to the inter-device distance, and then propose Dynamic Clustering and Partial-Steady Grouping algorithms that partition FL participants to achieve near-IIDness in their dataset while considering device mobility. Our mechanism significantly outperforms benchmark grouping algorithms at least by 110 times in terms of the joint cost between the number of dropout devices and the evenness in per-group device count, with a mild increase in the number of groups only by up to 0.93 groups.

cs.AI↗

Toward Scalable and Efficient Visual Data Transmission in 6G Networks

6G network technology will emerge in a landscape where visual data transmissions dominate global mobile traffic and are expected to grow continuously, driven by the increasing demand for AI-based computer vision applications. This will make already challenging task of visual data transmission even more difficult. In this work, we review effective techniques for visual data transmission, such as content compression and adaptive video streaming, highlighting their advantages and limitations. Further, considering the scalability and cost issues of cloud-based and on-device AI services, we explore distributed in-network computing architecture like fog-computing as a direction of 6G networks, and investigate the necessary technical properties for the timely delivery of visual data.

cs.NI↗

A Learning-based Distributed Algorithm for Scheduling in Multi-hop Wireless Networks

We address the joint problem of learning and scheduling in multi-hop wireless network without a prior knowledge on link rates. Previous scheduling algorithms need the link rate information, and learning algorithms often require a centralized entity and polynomial complexity. These become a major obstacle to develop an efficient learning-based distributed scheme for resource allocation in large-scale multi-hop networks. In this work, by incorporating with learning algorithm, we develop provably efficient scheduling scheme under packet arrival dynamics without a priori link rate information. We extend the results to distributed implementation and evaluation their performance through simulations.

cs.NI↗

CycleGANAS: Differentiable Neural Architecture Search for CycleGAN

We develop a Neural Architecture Search (NAS) framework for CycleGAN that carries out unpaired image-to-image translation task. Extending previous NAS techniques for Generative Adversarial Networks (GANs) to CycleGAN is not straightforward due to the task difference and greater search space. We design architectures that consist of a stack of simple ResNet-based cells and develop a search method that effectively explore the large search space. We show that our framework, called CycleGANAS, not only effectively discovers high-performance architectures that either match or surpass the performance of the original CycleGAN, but also successfully address the data imbalance by individual architecture search for each translation direction. To our best knowledge, it is the first NAS result for CycleGAN and shed light on NAS for more complex structures.

cs.CV↗

Delay-Based Back-Pressure Scheduling in Multihop Wireless Networks

Scheduling is a critical and challenging resource allocation mechanism for multihop wireless networks. It is well known that scheduling schemes that favor links with larger queue length can achieve high throughput performance. However, these queue-length-based schemes could potentially suffer from large (even infinite) packet delays due to the well-known last packet problem, whereby packets belonging to some flows may be excessively delayed due to lack of subsequent packet arrivals. Delay-based schemes have the potential to resolve this last packet problem by scheduling the link based on the delay the packet has encountered. However, characterizing throughput-optimality of these delay-based schemes has largely been an open problem in multihop wireless networks (except in limited cases where the traffic is single-hop.) In this paper, we investigate delay-based scheduling schemes for multihop traffic scenarios with fixed routes. We develop a scheduling scheme based on a new delay metric, and show that the proposed scheme achieves optimal throughput performance. Further, we conduct simulations to support our analytical results, and show that the delay-based scheduler successfully removes excessive packet delays, while it achieves the same throughput region as the queue-length-based scheme.

cs.NI↗

Throughput-optimal Scheduling in Multi-hop Wireless Networks without Per-flow Information

In this paper, we consider the problem of link scheduling in multi-hop wireless networks under general interference constraints. Our goal is to design scheduling schemes that do not use per-flow or per-destination information, maintain a single data queue for each link, and exploit only local information, while guaranteeing throughput optimality. Although the celebrated back-pressure algorithm maximizes throughput, it requires per-flow or per-destination information. It is usually difficult to obtain and maintain this type of information, especially in large networks, where there are numerous flows. Also, the back-pressure algorithm maintains a complex data structure at each node, keeps exchanging queue length information among neighboring nodes, and commonly results in poor delay performance. In this paper, we propose scheduling schemes that can circumvent these drawbacks and guarantee throughput optimality. These schemes use either the readily available hop-count information or only the local information for each link. We rigorously analyze the performance of the proposed schemes using fluid limit techniques via an inductive argument and show that they are throughput-optimal. We also conduct simulations to validate our theoretical results in various settings, and show that the proposed schemes can substantially improve the delay performance in most scenarios.

cs.NI↗