arXiv subjects

Yeseong Kim

Publications and source records attributed to Yeseong Kim.

14 recordsLinked to original sources

arXiv2026

FOCUS & RePAIR: Mitigating Text Degeneration via Token-Level Guidance for Pruned Large Language Models

Pruning is a practical approach to compress large language models (LLMs), but it can amplify text degeneration, especially repetition loops, even when perplexity and task accuracy remain largely unchanged. In this work, we present a token-level analysis of this failure mode by viewing decoding as a dynamical process that enters and persists in a small set of recurrent contexts. Our analysis decomposes degeneration into loop entry risk and loop persistence, and shows that persistence is controlled by the escape mass assigned to plausible alternatives within the token sampling set. Motivated by these findings, we propose two token-level guidance objectives for post-pruning fine-tuning. FOCUS reweights distillation toward high-confidence teacher regions to suppress leakage, while RePAIR uses onset-centered positive/negative continuation pairs with a margin loss to promote plausible alternatives and prevent early commitment to repetition loops. Experiments on open-ended continuation and instruction-based generation show that both methods consistently reduce repetition and improve generation quality.

cs.CL↗

arXiv2026

GlowQ: Group-Shared LOw-Rank Approximation for Quantized LLMs

Quantization techniques such as BitsAndBytes, AWQ, and GPTQ are widely used as a standard method in deploying large language models but often degrades accuracy when using low-bit representations, e.g., 4 bits. Low-rank correction methods (e.g., LQER, QERA, ASER) has been proposed to mitigate this issue, however, they restore all layers and insert error-correction modules into every decoder block, which increases latency and memory overhead. To address this limitation, we propose GlowQ, a group-shared low-rank approximation for quantized LLMs that caches a single shared right factor per input-sharing group and restores only the groups or layers that yield the highest accuracy benefit. GlowQ computes the high-precision projection once per input-sharing group and reuses it across its modules, reducing parameter and memory overhead, and retaining the expressivity of layer-specific corrections. We also propose a selective variant, GlowQ-S, that applies the cached shared module only where it provides the largest benefit. Compared with strong baselines, our approach reduces TTFB by (5.6%) and increases throughput by (9.6%) on average, while reducing perplexity on WikiText-2 by (0.17%) and increasing downstream accuracy by 0.42 percentage points. The selective model GlowQ-S further reduces latency, cutting TTFB by (23.4%) and increasing throughput by (37.4%), while maintaining accuracy within 0.2 percentage points on average. Code is available at https://github.com/ahnselim/GlowQ.

cs.LG↗

arXiv2026

FineScope : SAE-guided Data Selection Enables Domain Specific LLM Pruning and Finetuning

Training large language models (LLMs) from scratch requires significant computational resources, driving interest in developing smaller, domain-specific LLMs that maintain both efficiency and strong task performance. Medium-sized models such as LLaMA, llama} have served as starting points for domain-specific adaptation, but they often suffer from accuracy degradation when tested on specialized datasets. We introduce FineScope, a framework for deriving compact, domain-optimized LLMs from larger pretrained models. FineScope leverages the Sparse Autoencoder (SAE) framework, inspired by its ability to produce interpretable feature representations, to extract domain-specific subsets from large datasets. We apply structured pruning with domain-specific constraints, ensuring that the resulting pruned models retain essential knowledge for the target domain. To further enhance performance, these pruned models undergo self-data distillation, leveraging SAE-curated datasets to restore key domain-specific information lost during pruning. Extensive experiments and ablation studies demonstrate that FineScope achieves highly competitive performance, outperforming several large-scale state-of-the-art LLMs in domain-specific tasks. Additionally, our results show that FineScope enables pruned models to regain a substantial portion of their original performance when fine-tuned with SAE-curated datasets. Furthermore, applying these datasets to fine-tune pretrained LLMs without pruning also improves their domain-specific accuracy, highlighting the robustness of our approach.

cs.CL↗

arXiv2025

Flexible In-NAND Cryptographic Processing for Secure Flash Storage

We present FlashVault, an in-NAND self-encryption architecture that embeds a reconfigurable cryptographic engine into the unused silicon area of a state-of-the-art 4D V-NAND structure. FlashVault supports not only block ciphers for data encryption but also public-key and post-quantum algorithms for digital signatures, all within the NAND flash chip. This design enables each NAND chip to operate as a self-contained enclave without incurring area overhead, while eliminating the need for off-chip encryption. We implement FlashVault at the register-transfer level (RTL) and perform place-and-route (P&R) for accurate power/area evaluation. Our analysis shows that the power budget determines the number of cryptographic engines per NAND chip. We integrate this architectural choice into a full-system simulation and evaluate its performance on a wide range of cryptographic algorithms. Our results show that FlashVault consistently outperforms both CPU-based encryption (1.46~3.45x) and near-core processing architecture (1.02~2.01x), demonstrating its effectiveness as a secure SSD architecture that meets diverse cryptographic requirements imposed by regulatory standards and enterprise policies.

cs.AR↗

arXiv2025

A Diffusion-Based Framework for Configurable and Realistic Multi-Storage Trace Generation

We propose DiTTO, a novel diffusion-based framework for generating realistic, precisely configurable, and diverse multi-device storage traces. Leveraging advanced diffusion techniques, DiTTO enables the synthesis of high-fidelity continuous traces that capture temporal dynamics and inter-device dependencies with user-defined configurations. Our experimental results demonstrate that DiTTO can generate traces with high fidelity and diversity while aligning closely with guided configurations with only 8% errors.

cs.CV↗

arXiv2025

Rethinking Caching for LLM Serving Systems: Beyond Traditional Heuristics

Serving Large Language Models (LLMs) at scale requires meeting strict Service Level Objectives (SLOs) under severe computational and memory constraints. Nevertheless, traditional caching strategies fall short: exact-matching and prefix caches neglect query semantics, while state-of-the-art semantic caches remain confined to traditional intuitions, offering little conceptual departure. Building on this, we present SISO, a semantic caching system that redefines efficiency for LLM serving. SISO introduces centroid-based caching to maximize coverage with minimal memory, locality-aware replacement to preserve high-value entries, and dynamic thresholding to balance accuracy and latency under varying workloads. Across diverse datasets, SISO delivers up to 1.71$\times$ higher hit ratios and consistently stronger SLO attainment compared to state-of-the-art systems.

cs.DB↗

arXiv2025

Jack Unit: An Area- and Energy-Efficient Multiply-Accumulate (MAC) Unit Supporting Diverse Data Formats

In this work, we introduce an area- and energy-efficient multiply-accumulate (MAC) unit, named Jack unit, that is a jack-of-all-trades, supporting various data formats such as integer (INT), floating point (FP), and microscaling data format (MX). It provides bit-level flexibility and enhances hardware efficiency by i) replacing the carry-save multiplier (CSM) in the FP multiplier with a precision-scalable CSM, ii) performing the adjustment of significands based on the exponent differences within the CSM, and iii) utilizing 2D sub-word parallelism. To assess effectiveness, we implemented the layout of the Jack unit and three baseline MAC units. Additionally, we designed an AI accelerator equipped with our Jack units to compare with a state-of-the-art AI accelerator supporting various data formats. The proposed MAC unit occupies 1.17~2.01x smaller area and consumes 1.05~1.84x lower power compared to the baseline MAC units. On five AI benchmarks, the accelerator designed with our Jack units improves energy efficiency by 1.32~5.41x over the baseline across various data formats.

cs.AR↗

arXiv2025

FlexNeRFer: A Multi-Dataflow, Adaptive Sparsity-Aware Accelerator for On-Device NeRF Rendering

Neural Radiance Fields (NeRF), an AI-driven approach for 3D view reconstruction, has demonstrated impressive performance, sparking active research across fields. As a result, a range of advanced NeRF models has emerged, leading on-device applications to increasingly adopt NeRF for highly realistic scene reconstructions. With the advent of diverse NeRF models, NeRF-based applications leverage a variety of NeRF frameworks, creating the need for hardware capable of efficiently supporting these models. However, GPUs fail to meet the performance, power, and area (PPA) cost demanded by these on-device applications, or are specialized for specific NeRF algorithms, resulting in lower efficiency when applied to other NeRF models. To address this limitation, in this work, we introduce FlexNeRFer, an energy-efficient versatile NeRF accelerator. The key components enabling the enhancement of FlexNeRFer include: i) a flexible network-on-chip (NoC) supporting multi-dataflow and sparsity on precision-scalable MAC array, and ii) efficient data storage using an optimal sparsity format based on the sparsity ratio and precision modes. To evaluate the effectiveness of FlexNeRFer, we performed a layout implementation using 28nm CMOS technology. Our evaluation shows that FlexNeRFer achieves 8.2~243.3x speedup and 24.1~520.3x improvement in energy efficiency over a GPU (i.e., NVIDIA RTX 2080 Ti), while demonstrating 4.2~86.9x speedup and 2.3~47.5x improvement in energy efficiency compared to a state-of-the-art NeRF accelerator (i.e., NeuRex).

cs.AR↗

arXiv2025

Exploiting Boosting in Hyperdimensional Computing for Enhanced Reliability in Healthcare

Hyperdimensional computing (HDC) enables efficient data encoding and processing in high-dimensional space, benefiting machine learning and data analysis. However, underutilization of these spaces can lead to overfitting and reduced model reliability, especially in data-limited systems a critical issue in sectors like healthcare that demand robustness and consistent performance. We introduce BoostHD, an approach that applies boosting algorithms to partition the hyperdimensional space into subspaces, creating an ensemble of weak learners. By integrating boosting with HDC, BoostHD enhances performance and reliability beyond existing HDC methods. Our analysis highlights the importance of efficient utilization of hyperdimensional spaces for improved model performance. Experiments on healthcare datasets show that BoostHD outperforms state-of-the-art methods. On the WESAD dataset, it achieved an accuracy of 98.37%, surpassing Random Forest, XGBoost, and OnlineHD. BoostHD also demonstrated superior inference efficiency and stability, maintaining high accuracy under data imbalance and noise. In person-specific evaluations, it achieved an average accuracy of 96.19%, outperforming other models. By addressing the limitations of both boosting and HDC, BoostHD expands the applicability of HDC in critical domains where reliability and precision are paramount.

cs.LG↗

arXiv2023

Efficient Off-Policy Reinforcement Learning via Brain-Inspired Computing

Reinforcement Learning (RL) has opened up new opportunities to enhance existing smart systems that generally include a complex decision-making process. However, modern RL algorithms, e.g., Deep Q-Networks (DQN), are based on deep neural networks, resulting in high computational costs. In this paper, we propose QHD, an off-policy value-based Hyperdimensional Reinforcement Learning, that mimics brain properties toward robust and real-time learning. QHD relies on a lightweight brain-inspired model to learn an optimal policy in an unknown environment. On both desktop and power-limited embedded platforms, QHD achieves significantly better overall efficiency than DQN while providing higher or comparable rewards. QHD is also suitable for highly-efficient reinforcement learning with great potential for online and real-time learning. Our solution supports a small experience replay batch size that provides 12.3 times speedup compared to DQN while ensuring minimal quality loss. Our evaluation shows QHD capability for real-time learning, providing 34.6 times speedup and significantly better quality of learning than DQN.

cs.LG↗

arXiv2022

DeepSketch: A New Machine Learning-Based Reference Search Technique for Post-Deduplication Delta Compression

Data reduction in storage systems is becoming increasingly important as an effective solution to minimize the management cost of a data center. To maximize data-reduction efficiency, existing post-deduplication delta-compression techniques perform delta compression along with traditional data deduplication and lossless compression. Unfortunately, we observe that existing techniques achieve significantly lower data-reduction ratios than the optimal due to their limited accuracy in identifying similar data blocks. In this paper, we propose DeepSketch, a new reference search technique for post-deduplication delta compression that leverages the learning-to-hash method to achieve higher accuracy in reference search for delta compression, thereby improving data-reduction efficiency. DeepSketch uses a deep neural network to extract a data block's sketch, i.e., to create an approximate data signature of the block that can preserve similarity with other blocks. Our evaluation using eleven real-world workloads shows that DeepSketch improves the data-reduction ratio by up to 33% (21% on average) over a state-of-the-art post-deduplication delta-compression technique.

cs.LG↗

arXiv2021

Spiking Hyperdimensional Network: Neuromorphic Models Integrated with Memory-Inspired Framework

Recently, brain-inspired computing models have shown great potential to outperform today's deep learning solutions in terms of robustness and energy efficiency. Particularly, Spiking Neural Networks (SNNs) and HyperDimensional Computing (HDC) have shown promising results in enabling efficient and robust cognitive learning. Despite the success, these two brain-inspired models have different strengths. While SNN mimics the physical properties of the human brain, HDC models the brain on a more abstract and functional level. Their design philosophies demonstrate complementary patterns that motivate their combination. With the help of the classical psychological model on memory, we propose SpikeHD, the first framework that fundamentally combines Spiking neural network and hyperdimensional computing. SpikeHD generates a scalable and strong cognitive learning system that better mimics brain functionality. SpikeHD exploits spiking neural networks to extract low-level features by preserving the spatial and temporal correlation of raw event-based spike data. Then, it utilizes HDC to operate over SNN output by mapping the signal into high-dimensional space, learning the abstract information, and classifying the data. Our extensive evaluation on a set of benchmark classification problems shows that SpikeHD provides the following benefit compared to SNN architecture: (1) significantly enhance learning capability by exploiting two-stage information processing, (2) enables substantial robustness to noise and failure, and (3) reduces the network size and required parameters to learn complex information.

cs.NE↗

arXiv2020

SHEARer: Highly-Efficient Hyperdimensional Computing by Software-Hardware Enabled Multifold Approximation

Hyperdimensional computing (HD) is an emerging paradigm for machine learning based on the evidence that the brain computes on high-dimensional, distributed, representations of data. The main operation of HD is encoding, which transfers the input data to hyperspace by mapping each input feature to a hypervector, accompanied by so-called bundling procedure that simply adds up the hypervectors to realize encoding hypervector. Although the operations of HD are highly parallelizable, the massive number of operations hampers the efficiency of HD in embedded domain. In this paper, we propose SHEARer, an algorithm-hardware co-optimization to improve the performance and energy consumption of HD computing. We gain insight from a prudent scheme of approximating the hypervectors that, thanks to inherent error resiliency of HD, has minimal impact on accuracy while provides high prospect for hardware optimization. In contrast to previous works that generate the encoding hypervectors in full precision and then ex-post quantizing, we compute the encoding hypervectors in an approximate manner that saves a significant amount of resources yet affords high accuracy. We also propose a novel FPGA implementation that achieves striking performance through massive parallelism with low power consumption. Moreover, we develop a software framework that enables training HD models by emulating the proposed approximate encodings. The FPGA implementation of SHEARer achieves an average throughput boost of 104,904x (15.7x) and energy savings of up to 56,044x (301x) compared to state-of-the-art encoding methods implemented on Raspberry Pi 3 (GeForce GTX 1080 Ti) using practical machine learning datasets.

cs.LG↗

arXiv2019

RAPIDNN: In-Memory Deep Neural Network Acceleration Framework

Deep neural networks (DNN) have demonstrated effectiveness for various applications such as image processing, video segmentation, and speech recognition. Running state-of-the-art DNNs on current systems mostly relies on either generalpurpose processors, ASIC designs, or FPGA accelerators, all of which suffer from data movements due to the limited onchip memory and data transfer bandwidth. In this work, we propose a novel framework, called RAPIDNN, which processes all DNN operations within the memory to minimize the cost of data movement. To enable in-memory processing, RAPIDNN reinterprets a DNN model and maps it into a specialized accelerator, which is designed using non-volatile memory blocks that model four fundamental DNN operations, i.e., multiplication, addition, activation functions, and pooling. The framework extracts representative operands of a DNN model, e.g., weights and input values, using clustering methods to optimize the model for in-memory processing. Then, it maps the extracted operands and their precomputed results into the accelerator memory blocks. At runtime, the accelerator identifies computation results based on efficient in-memory search capability which also provides tunability of approximation to further improve computation efficiency. Our evaluation shows that RAPIDNN achieves 68.4x, 49.5x energy efficiency improvement and 48.1x, 10.9x speedup as compared to ISAAC and PipeLayer, the state-of-the-art DNN accelerators, while ensuring less than 0.3% of quality loss.

cs.NE↗