Searcharxiv⌕ Search

arXiv subjects

Hyunwoo Oh

Publications and source records attributed to Hyunwoo Oh.

At least 19 recordsLinked to original sources
arXiv2026

Asymmetric Capacity Allocation in Self-Refinement Pipelines

Self-refinement, typically structured as generation, critique, and revision, is a widely adopted paradigm for improving LLM generation and serves as a core mechanism in many LLM agents. While the three stages involve different cognitive demands, most existing approaches conveniently treat the model size as an implementation detail rather than a subject of study, which may lead to a waste of resources. Little work has systematically examined how model size affects each stage or whether effective self-refinement requires equally capable models for generation, critique, and revision. We present the first stage-wise model size study of the self-refinement pipeline on 5 benchmarks from different domains using 6 model sizes of Qwen3 and 4 model sizes of Gemma 3. We conclude that larger generators and refiners generally improve the pipeline, whereas an undersized refiner can even harm performance. Second, performance is highly insensitive to the size of the critic, although including even a small critic consistently outperforms omitting critique altogether. Our findings demonstrate that model capacity should not be allocated uniformly across self-refinement pipelines. Instead, different stages exhibit distinct size scaling characteristics, providing practical guidance for designing more computationally efficient multi-stage language model systems.

cs.LG↗
arXiv2026

Vector Symbolic Policy Gradient

We answer this question with Vector-Symbolic Policy Gradient (VSPG), a discrete-action actor that represents each action by a unit-norm hypervector and scores it by similarity to the encoded state. Under the standard softmax policy-gradient surrogate, we prove that its update is exactly advantage-weighted hypervector bundling followed by normalization, and therefore supports standard advantage estimators. We further show that each trained action hypervector is a fixed-size compressed kernel memory, storing an advantage-weighted kernel expansion over visited states and transferring evidence according to the encoder-induced similarity. This provides a concrete mechanism that can support sample-efficient learning without increasing inference-time memory. Finally, for bipolar action memories, we prove that greedy action selection is stable under random bit flips, with failure probability decaying exponentially in the hypervector dimension. VSPG thus connects VSA action memories, log-linear policy gradients, and kernel policy search while providing a quantitative robustness guarantee.

cs.LG↗
arXiv2026

Trident : How to Break Deep Reinforcement Learning Cyber Defenses (Agentic)

Autonomous cyber defense systems based on Deep Reinforcement Learning (DRL) have attracted significant research attention, yet remain evaluated almost exclusively against static, heuristic red agents, leaving their robustness against adaptive threats critically understudied. Meanwhile, recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) have improved LLM reasoning, but their integration into cybersecurity remains elusive due to the absence of suitable benchmark environments and interaction datasets. To bridge this gap, we introduce Trident, an agentic LLM red teaming framework comprising three components: a dynamic benchmark with isolated sandbox servers spanning CybORG CAGE 4 and CyberWheel, a dataset comprises over 13,000 high-fidelity red-blue interaction trajectories for RLVR, and a ``Code-as-Policy'' RLVR agentic architecture Trident Agentic). The latter reformulates red agent training as a contextual bandit via a tripartite Log Summarizer--Planner--Coder design, where a trainable Planner generates complete attack strategies from compressed execution logs, which a frozen Coder translates into executable Python policies deployed against live DRL defenders. Empirical evaluations reveal a fundamental brittleness in existing defenses: with a single trainable 7B planner, Trident reduces blue agent defensive performance by an average of 522% compared to static red agent baselines while autonomously discovering emergent behaviors such as decoy avoidance and adaptive state prioritization that static heuristics entirely fail to uncover.

cs.CR↗
arXiv2026

PolyQ: Codesigning End-to-End Quantization Framework for Scalable Edge CPU LLM Inference

CPUs are the most universal target for on-device LLM inference, but existing low-bit quantization methods offer either coarse operating points or fine-grained mixed precision that is difficult to execute efficiently on CPUs. We present PolyQ, a CPU-oriented compiler/quantization co-design for activation-aware channel-wise bit allocation under a user-specified average-bit budget. PolyQ assigns per-channel bit-widths from $\{2,3,4,8,16\}$, then uses a compile-time model compiler to permute and cluster channels into bit-homogeneous blocks, generate SIMD- and LUT-compatible kernels, and merge compatible permutations across operators to keep layout regularization off the runtime path. This turns fine-grained budget fitting into a practical fractional-bit deployment method for CPU-only inference. Across Falcon-H1-3B, Llama2-13B, and Qwen3-32B on WikiText-2, PolyQ provides stable quality scaling from 3--6\,b and improves perplexity by 2.4--32.1\% over prior methods at a 3\,b target. End-to-end measurements on three representative CPUs -- workstation, laptop, and mobile -- show that compiler layout regularization reduces activation reorder traffic by up to 70.8\%, prefill latency and decode throughput scale nearly proportionally with the configured bit budget, and energy/token overhead stays below 2\% relative to an optimized LUT-based back-end. These results show that fractional-bit CPU deployment is practical, predictable, and energy-efficient across diverse edge targets.

cs.LG↗
arXiv2026

ExaGEMM: Exploration Framework for CPU-Driven ML Inference via Associative In-Register Computing for Low-Bit GEMM

Low-bit GEMM is increasingly central to efficient ML inference, yet very-low-bit execution remains a poor fit for conventional CPUs. Practical deployment spans fragmented regimes-from 1/2/4-bit weights to varying activation precision-whose feasibility, reuse opportunity, and support cost differ under fixed SIMD and register-file budgets, making lightweight CPU support selection a first-class design problem. We present ExaGEMM, a workload-aware codesign and exploration framework for CPU-native low-bit GEMM via register-resident LUT execution. The key insight is that existing SIMD datapaths already cover table generation and accumulation; the only new hardware is an in-register select/feed mechanism with explicitly modeled cost. ExaGEMM co-explores parameterized kernels and lightweight SIMD ISA support using analytical models of register feasibility, compute cost, memory traffic, and hardware overhead, pruning the candidate space by 99.2% before simulation. It then identifies non-dominated support points and generates ISA specs, gem5 patches, and GEMM kernels for validation. Across representative ML models and CPU targets, ExaGEMM improves latency by 13.29x over software-only baselines, while showing that workload-aware frontier selection is especially important for mixed-precision LLM workloads.

cs.AR↗
arXiv2026

Qubit-Efficient Quantum Search for Hyperdimensional Decomposition via Logarithmic Encoding

Hyperdimensional Computing (HDC) represents symbols using high-dimensional hypervectors of dimension $D$. In hypervector decomposition, the objective is to recover $F$ constituent hypervectors, each drawn from a codebook of size $N$, from a bound target hypervector. This requires searching over $N^F$ candidate tuples, making the task computationally prohibitive at scale. Recent quantum approach provides a quadratic search advantage, but typically rely on qubit-inefficient $O(D)$-qubit hypervector representations. We propose a qubit-efficient quantum framework for HDC decomposition that reduces the representation cost to $O(\log D)$. The framework introduces logarithmic hypervector and binding encodings, together with a reversible hypervector lookup operator for circuit-level manipulation of dense hypervectors. Combined with a modified D\"urr-H{\o}yer search procedure, the method preserves $O(\sqrt{N^F})$ search complexity while substantially reducing qubit usage. Experimental results validate correct similarity computation, accurate decomposition in executable regimes, and significantly improved qubit scaling over baselines based on explicit $D$-qubit hypervector encodings, achieving up to $2{,}000\times$ fewer qubits.

cs.LG↗
arXiv2026

FusionSense: Tri-Stage Near-Sensor Learning for Runtime-Adaptive Multimodal Edge Intelligence

Autonomous systems and smart-industry deployments increasingly split computation across near-sensor, edge, and cloud resources, where tight energy, latency, and reliability budgets demand run-time adaptivity. In practice, deciding what to compute and transmit at each point is pivotal; yet as multimodal sensor suites (cameras, LiDAR/depth, etc.) proliferate at the edge, most prior approaches either (i) fuse modalities on powerful servers or (ii) apply uni-modal near-sensor filters that ignore cross-modal dependencies, leading to redundant transmissions or missed events. We present FusionSense, a fusion-aware intelligent sensing framework for energy-constrained autonomous edge systems. Lightweight near-sensor classifiers are trained via a three-step procedure: (i) a server-side fusion model learns the downstream task, (ii) filter-out-safe (FoS) labels quantify each modality's necessity relative to the fused decision, and (iii) an edge-side fusion model is compacted by injecting near-sensor predictions as auxiliary signals. The result is a run-time decision layer that jointly reduces compute and communication while scaling linearly with sensor count. On a dual-modality (RGB+Depth/LiDAR) setup with SynDrone, FusionSense sustains task quality at substantially higher data-reduction rates than uni-modal filters and delivers large end-to-end gains: up to 33x lower energy at 1% FoI prevalence, 11x at 10%, a 92.3% reduction in quality loss at a fixed 30% data reduction, and roughly 1.5x higher energy savings than the best prior filtering baseline.

cs.LG↗
arXiv2026

TorR: Towards Brain-Inspired Task-Oriented Reasoning via Cache-Oriented Algorithm-Architecture Co-design

Task-oriented object detection (TOOD) atop CLIP offers open-vocabulary, prompt-driven semantics, yet dense per-window computation and heavy memory traffic hinder real-time, power-limited edge deployment. We present \emph{TorR}, a brain-inspired \textbf{algorithm--architecture co-design} that \textbf{replaces CLIP-style dense alignment with a hyperdimensional (HDC) associative reasoner} and turns temporal coherence into reuse. On the \emph{algorithm} side, TorR reformulates alignment as HDC similarity and graph composition, introducing \emph{partial-similarity reuse} via (i) query caching with per-class score accumulation, (ii) exact $\delta$-updates when only a small set of hypervector bits change, and (iii) similarity/load-gated bypass under high system load. On the \emph{architecture} side, TorR instantiates a lane-scalable, bit-sliced item memory with bank/precision gating and a lightweight controller that schedules bypass/$\delta$/full paths to meet RT-30/RT-60 targets as object counts vary. Synthesized in a TSMC 28\,nm process and exercised with a cycle-accurate simulator, TorR sustains real-time throughput with millijoule-scale energy per window ($\approx$50\,mJ at 60\,FPS; $\approx$113\,mJ at 30\,FPS) and low latency jitter, while delivering competitive AP@0.5 across five task prompts (mean 44.27\%) within a bounded margin to strong VLM baselines, but at orders-of-magnitude lower energy. The design exposes deployment-time configurability (effective dimension $D'$, thresholds, precision) to trade accuracy, latency, and energy for edge budgets.

cs.AR↗
arXiv2026

TRINE: A Token-Aware, Runtime-Adaptive FPGA Inference Engine for Multimodal AI

Multimodal stacks that mix ViTs, CNNs, GNNs, and transformer NLP strain embedded platforms because their compute/memory patterns diverge and hard real-time targets leave little slack. TRINE is a single-bitstream FPGA accelerator and compiler that executes end-to-end multimodal inference without reconfiguration. Layers are unified as DDMM/SDDMM/SpMM and mapped to a mode-switchable engine that toggles at runtime among weight/output-stationary systolic, 1xCS SIMD, and a routable adder tree (RADT) on a shared PE array. A width-matched, two-stage top-k unit enables in-stream token pruning, while dependency-aware layer offloading (DALO) overlaps independent kernels across reconfigurable processing units to sustain utilization. Evaluated on Alveo U50 and ZCU104, TRINE reduces latency by up to 22.57x vs. RTX 4090 and 6.86x vs. Jetson Orin Nano at 20-21 W; token pruning alone yields up to 7.8x on ViT-heavy pipelines, and DALO contributes up to 79% throughput improvement. With int8 quantization, accuracy drops remain <2.5% across representative tasks, delivering state-of-the-art latency and energy efficiency for unified vision, language, and graph workloads-in one bitstream.

cs.AR↗
arXiv2026

ADAPT: Attention Driven Adaptive Prompt Scheduling and InTerpolating Orthogonal Complements for Rare Concepts Generation

Generating rare compositional concepts in text-to-image synthesis remains a challenge for diffusion models, particularly for attributes that are uncommon in the training data. While recent approaches, such as R2F, address this challenge by utilizing LLM for prompt scheduling, they suffer from inherent variance due to the randomness of language models and suboptimal guidance from iterative text embedding switching. To address these problems, we propose the ADAPT framework, a training-free framework that deterministically plans and semantically aligns prompt schedules, providing consistent guidance to enhance the composition of rare concepts. By leveraging attention scores and orthogonal components, ADAPT significantly enhances compositional generation of rare concepts in the RareBench benchmark without additional training or fine-tuning. Through comprehensive experiments, we demonstrate that ADAPT achieves superior performance in RareBench and accurately reflects the semantic information of rare attributes, providing deterministic and precise control over the generation of rare compositions without compromising visual integrity.

cs.CV↗
arXiv2026

Adaptive Auxiliary Prompt Blending for Target-Faithful Diffusion Generation

Diffusion-based text-to-image (T2I) models have made remarkable progress in generating photorealistic and semantically rich images. However, when the target concepts lie in low-density regions of the training distribution, these models often produce semantically misaligned or structurally inconsistent results. This limitation arises from the long-tailed nature of text-image datasets, where rare concepts or editing instructions are underrepresented. To address this, we introduce Adaptive Auxiliary Prompt Blending (AAPB) - a unified framework that stabilizes the diffusion process in low-density regions. AAPB leverages auxiliary anchor prompts to provide semantic support in rare concept generation and structural support in image editing, ensuring faithful guidance toward the target prompt. Unlike prior heuristic prompt alternation methods, AAPB derives a closed-form adaptive coefficient that optimally balances the influence between the auxiliary anchor and the target prompt at each diffusion step. Grounded in Tweedie's identity, our formulation provides a principled and training-free framework for adaptive prompt blending, ensuring stable and target-faithful generation. We demonstrate the effectiveness of adaptive interpolation over fixed interpolation through controlled experiments and empirically show consistent improvements on the RareBench and FlowEdit datasets, achieving superior semantic accuracy and structural fidelity compared to prior training-free baselines.

cs.CV↗
arXiv2026

Follow the Saliency: Supervised Saliency for Retrieval-augmented Dense Video Captioning

Existing retrieval-augmented approaches for Dense Video Captioning (DVC) often fail to achieve accurate temporal segmentation aligned with true event boundaries, as they rely on heuristic strategies that overlook ground truth event boundaries. The proposed framework, \textbf{STaRC}, overcomes this limitation by supervising frame-level saliency through a highlight detection module. Note that the highlight detection module is trained on binary labels derived directly from DVC ground truth annotations without the need for additional annotation. We also propose to utilize the saliency scores as a unified temporal signal that drives retrieval via saliency-guided segmentation and informs caption generation through explicit Saliency Prompts injected into the decoder. By enforcing saliency-constrained segmentation, our method produces temporally coherent segments that align closely with actual event transitions, leading to more accurate retrieval and contextually grounded caption generation. We conduct comprehensive evaluations on the YouCook2 and ViTT benchmarks, where STaRC achieves state-of-the-art performance across most of the metrics. Our code is available at https://github.com/ermitaju1/STaRC

cs.CV↗
arXiv2026

Conditional neural control variates for variance reduction in Bayesian inverse problems

Bayesian inference for inverse problems involves computing expectations under posterior distributions--e.g., posterior means, variances, or predictive quantities--typically via Monte Carlo (MC) estimation. When the quantity of interest varies significantly under the posterior, accurate estimates demand many samples--a cost often prohibitive for partial differential equation-constrained problems. To address this challenge, we introduce conditional neural control variates, a modular method that learns amortized control variates from joint model-data samples to reduce the variance of MC estimators. To scale to high-dimensional problems, we leverage Stein's identity to design an architecture based on an ensemble of hierarchical coupling layers with tractable Jacobian trace computation. Training requires: (i) samples from the joint distribution of unknown parameters and observed data; and (ii) the posterior score function, which can be computed from physics-based likelihood evaluations, neural operator surrogates, or learned generative models such as conditional normalizing flows. Once trained, the control variates generalize across observations without retraining. We validate our approach on stylized and partial differential equation-constrained Darcy flow inverse problems, outperforming classical Stein control variates and achieving substantial variance reduction, even when the analytical score is replaced by a learned surrogate.

stat.ML↗
arXiv2026

$n$-Musketeers: Reinforcement Learning Shapes Collaboration Among Language Models

Recent progress in reinforcement learning with verifiable rewards (RLVR) shows that small, specialized language models (SLMs) can exhibit structured reasoning without relying on large monolithic LLMs. We introduce soft hidden-state collaboration, where multiple heterogeneous frozen SLM experts are integrated through their internal representations via a trainable attention interface. Experiments on Reasoning Gym and GSM8K show that this latent integration is competitive with strong single-model RLVR baselines. Ablations further reveal a dual mechanism of expert utilization: for simpler arithmetic domains, performance gains can largely be explained by static expert preferences, whereas more challenging settings induce increasingly concentrated and structured expert attention over training, indicating emergent specialization in how the router connects to relevant experts. Overall, hidden-state collaboration provides a compact mechanism for leveraging frozen experts, while offering an observational window into expert utilization patterns and their evolution under RLVR.

cs.LG↗
arXiv2026

DecoHD: Decomposed Hyperdimensional Classification under Extreme Memory Budgets

Decomposition is a proven way to shrink deep networks without changing input-output dimensionality or interface semantics. We bring this idea to hyperdimensional computing (HDC), where footprint cuts usually shrink the feature axis and erode concentration and robustness. Prior HDC decompositions decode via fixed atomic hypervectors, which are ill-suited for compressing learned class prototypes. We introduce DecoHD, which learns directly in a decomposed HDC parameterization: a small, shared set of per-layer channels with multiplicative binding across layers and bundling at the end, yielding a large representational space from compact factors. DecoHD compresses along the class axis via a lightweight bundling head while preserving native bind-bundle-score; training is end-to-end, and inference remains pure HDC, aligning with in/near-memory accelerators. In evaluation, DecoHD attains extreme memory savings with only minor accuracy degradation under tight deployment budgets. On average it stays within about 0.1-0.15% of a strong non-reduced HDC baseline (worst case 5.7%), is more robust to random bit-flip noise, reaches its accuracy plateau with up to ~97% fewer trainable parameters, and--in hardware--delivers roughly 277x/35x energy/speed gains over a CPU (AMD Ryzen 9 9950X), 13.5x/3.7x over a GPU (NVIDIA RTX 4090), and 2.0x/2.4x over a baseline HDC ASIC.

cs.LG↗
arXiv2026

LogHD: Robust Compression of Hyperdimensional Classifiers via Logarithmic Class-Axis Reduction

Hyperdimensional computing (HDC) suits memory, energy, and reliability-constrained systems, yet the standard "one prototype per class" design requires $O(CD)$ memory (with $C$ classes and dimensionality $D$). Prior compaction reduces $D$ (feature axis), improving storage/compute but weakening robustness. We introduce LogHD, a logarithmic class-axis reduction that replaces the $C$ per-class prototypes with $n\!\approx\!\lceil\log_k C\rceil$ bundle hypervectors (alphabet size $k$) and decodes in an $n$-dimensional activation space, cutting memory to $O(D\log_k C)$ while preserving $D$. LogHD uses a capacity-aware codebook and profile-based decoding, and composes with feature-axis sparsification. Across datasets and injected bit flips, LogHD attains competitive accuracy with smaller models and higher resilience at matched memory. Under equal memory, it sustains target accuracy at roughly $2.5$-$3.0\times$ higher bit-flip rates than feature-axis compression; an ASIC instantiation delivers $498\times$ energy efficiency and $62.6\times$ speedup over an AMD Ryzen 9 9950X and $24.3\times$/$6.58\times$ over an NVIDIA RTX 4090, and is $4.06\times$ more energy-efficient and $2.19\times$ faster than a feature-axis HDC ASIC baseline.

cs.LG↗
arXiv2026

HYPERDOA: Robust and Efficient DoA Estimation using Hyperdimensional Computing

Direction of Arrival (DoA) estimation techniques face a critical trade-off, as classical methods often lack accuracy in challenging, low signal-to-noise ratio (SNR) conditions, while modern deep learning approaches are too energy-intensive and opaque for resource-constrained, safety-critical systems. We introduce HYPERDOA, a novel estimator leveraging Hyperdimensional Computing (HDC). The framework introduces two distinct feature extraction strategies -- Mean Spatial-Lag Autocorrelation and Spatial Smoothing -- for its HDC pipeline, and then reframes DoA estimation as a pattern recognition problem. This approach leverages HDC's inherent robustness to noise and its transparent algebraic operations to bypass the expensive matrix decompositions and "black-box" nature of classical and deep learning methods, respectively. Our evaluation demonstrates that HYPERDOA achieves ~35.39% higher accuracy than state-of-the-art methods in low-SNR, coherent-source scenarios. Crucially, it also consumes ~93% less energy than competing neural baselines on an embedded NVIDIA Jetson Xavier NX platform. This dual advantage in accuracy and efficiency establishes HYPERDOA as a robust and viable solution for mission-critical applications on edge devices.

eess.SP↗
arXiv2026

Numerical study of computational cost of maintaining adiabaticity for long paths

Recent work argued that the scaling of a dimensionless quantity $Q_D$ with path length is a better proxy for quantifying the scaling of the computational cost of maintaining adiabaticity than the timescale. It also conjectured that generically the scaling will be superlinear (although special cases exist in which it is linear). The quantity $Q_D$ depends only on the properties of ground states along the Hamiltonian path and the rate at which the path is followed. In this paper, we demonstrate that this conjecture holds for simple Hamiltonian systems that can be studied numerically. In particular, the systems studied exhibit the behavior that $Q_D$ grows approximately as $L \log L$ where $L$ is the path length when the threshold error is fixed.

quant-ph↗