arXiv subjects

Zijun Zhao

Publications and source records attributed to Zijun Zhao.

At least 19 recordsLinked to original sources

arXiv2026

MX-Like Tensor Core for LLM Inference with NPU Integration via Hardware-Software Co-Design Exploration

Block-scaled quantization formats such as OCP Microscaling (MX) and NVFP4 have been widely adopted as efficient data formats for LLM inference. However, the design space of these formats spanned by shared scale, weight, and activation format combinations has not been systematically explored, and the joint accuracy-silicon tradeoff across this space remains uncharacterized, leaving the true Pareto-optimal configurations for NPU-based edge LLM deployment an open question. To bridge this gap, we conduct a joint hardware-software (HW-SW) co-design exploration of the above underexplored MX-like design space with three innovations. 1) We define and systematically characterize the full 8-bit scale format family within the MX-like design space, jointly with asymmetric weight-activation precision, identifying configurations that outperform the MX and NVFP4-like (without per-tensor scaling, PTS) baselines. 2) We build a per-format specialized MAC datapath template that enables fair hardware cost evaluation across the space. 3) We develop an automated co-design flow, integrated into the open-source SNAX NPU platform, that enables joint HW-SW analysis across the design space. Overall, our co-design exploration reveals that scale format matters only under aggressive FP4 quantization, yielding a case-oriented design guideline for both scale format and asymmetric weight-activation precision. We further identify UE5M3 as a favorable accuracy-hardware tradeoff for the FP4 regime, reducing perplexity degradation by up to 25.3% relative to the NVFP4-like baseline, at a cost of only +0.06% area and +0.49% power overhead.

cs.AR↗

arXiv2026

GraphVQ: Structure-Aware Autoregressive Decoding over Context-Quantized Graph Tokens

Graph foundation models need a discrete token representation, but casting a graph as a generatable token sequence faces a structural obstacle: edges spanning beyond the serialization window cannot be emitted in one pass--so one-pass autoregressive generators systematically under-produce cycles--and a single global condition cannot tell candidate edges apart. GraphVQ removes both obstacles: node contexts--features plus a local edge mask under multi-order breadth-first serialization--are quantized into a shared codebook by a VQ-VAE with BCE-calibrated Bernoulli edge decoding, and a second-stage structure-aware decoder emits the global adjacency conditioned on token-derived pair features, whose necessity over any global-summary condition is formalized in a scoped impossibility result. The tokenizer reconstructs node features at 0.86--0.99 accuracy and decodes local edges at AUROC >= 0.89 (ECE <= 0.007). Under one same-split protocol on four datasets, pair conditioning improves orbit MMD 0.248 -> 0.174 on PROTEINS and 3.4x on a ring stress test, and vanishes on a random-label control--the signature of attribute--topology coupling--so the gain is claimed exactly where attributes carry edge-relevant signal. GraphVQ ranks first among learned generators on PROTEINS, ties for first on SYN-COMM, and improves orbit MMD 2.7--17x over one-stage generation on three datasets, with seed-level bootstrap intervals confirming the rankings are not seed noise; on MUTAG the unweighted edge target under-generates and is reported as such. These results locate the structural control of autoregressive graph generation in the granularity of the condition: pair-level token context turns a quantized vocabulary into a usable capacity axis for distribution-faithful graph generation and future token-level pretraining.

cs.LG↗

arXiv2026

RACE: Relation-Level Counterfactual Explanations for Heterogeneous Graph Neural Networks

Counterfactual explanations of graph neural networks identify edge deletions that flip a prediction. On heterogeneous graphs, however, existing methods first collapse the graph into untyped edges, so they cannot answer the question a domain expert actually asks: which relation type drives this prediction? We present RACE (Relation-Aware Counterfactual Explanations), which gives this question an exact, per-instance answer. For every explained instance, an exhaustive search over relation subsets returns the certified minimum relation-deletion set that flips the prediction -- or an explicit report that no such deletion exists; each relation-level answer is then refined into a typed edge set within the attributed relations, verified on the discrete model by single-edge restoration. The relation-level answer is exact and deterministic given the frozen backbone, whereas soft-mask baselines vary by 6-8 pp in success rate across runs differing only in random ordering. On ACM, a Cora-derived graph, and ogbn-mag, RACE improves counterfactual success rate over the strongest baseline by up to +2.7 pp while deleting fewer edges, and attains the highest success rate among all same-task baselines on every dataset; the advantage reproduces across four backbones on ogbn-arXiv and on DBLP, with cross-seed relation-set agreement up to 0.89. A synthetic study with known generating mechanisms confirms that the search recovers the relation the trained model actually relies on -- and reports infeasibility rather than fabricating an attribution when the model has learned none -- so the explanations stay trustworthy exactly where explanations matter.

cs.LG↗

arXiv2026

Where Privacy Belongs: Placement Diagnosis and Certified Selection for Private Counterfactual Explanations on Graphs

Counterfactual explanations for graph neural networks (GNNs) find the minimal intervention that flips a node's prediction--but computing one requires reading sensitive graph structure, and releasing it discloses that structure. Both existing placements fail. Privatizing the graph before explaining corrupts the target on exactly the borderline nodes needing recourse, manufacturing spurious flips that flip the privatized graph but not the true one. Explaining on the clean graph and perturbing the released explanation resists certification: re-auditing the standard heuristic shows an implied full-release budget of 573--753 on Cora and 256 on CiteSeer--orders of magnitude beyond its advertised budget--with worst-case single-entry leakage at AUC 1.0. We propose PrivCFS, which replaces certification-by-optimization with certification-by-construction: counterfactual selection over a fixed, data-independent candidate universe--edge interventions from a public prior graph, feature interventions from a public schema--whose no-op semantics give neighboring graphs the same output support. A validity-gated, clipped utility of global sensitivity $Δu \le 1$ released through the exponential mechanism gives pure $\varepsilon$-DP for the complete released object, composable over queries--to our knowledge the first such guarantee on graphs. Privacy noise is the cheapest stage: at $\varepsilon$=8 the release retains 94--97% of its support-restricted non-private optimum on the recourse population and 83--95% on the general one; the optimal edge-inference audit attains AUC 0.50 on average and 0.59 worst-pair, versus the heuristic's worst entry 1.0; and transfers to a 15K-node graph at 0.96 valid rate. The dominant cost is a measurable, monotone price in public disclosure, readable off one table before any budget is spent--turning explanation privacy from an accounting risk into a purchasable decision.

cs.LG↗

arXiv2026

Beyond the Beam: Constructive Repair and Candidate Completion for Generative Recommendation

Generative recommenders retrieve items by generating identifiers, but a valid identifier can remain outside the beam after catalog expansion. This raises two connected questions: which failures can identifier assignment repair, and how should retrieval proceed beyond the initial beam? We characterize assignment repair with a fixed generator and retained old identifiers. Output-invariance certificates identify failures shared by all admissible assignments. Under a common effective prefix, coupled support and ranking constraints give the exact feasible interval of new-item counts for target recovery. Building on this characterization, Beyond the Beam (BB) obtains minimum-replacement repairs through an integral flow formulation, selects a shared map and adapts the generator. At inference, generative likelihood and collaborative evidence define one score for ranking, candidate priority and stopping. Retained prefix bounds guide candidate completion and certify its global Top-$K$ when the stopping condition is met. Exhaustive finite-catalog evaluation confirms construction in every feasible case. Across three Amazon Reviews categories and three random seeds, the full T5 procedure improves mean Recall@10 by 15.5--46.3% and NDCG@10 by 15.2--44.4% over the best-performing evaluated generative baseline for each dataset and metric. Matched controls show that shared construction and adaptation improve new-target ranking and certification efficiency on Beauty and Toys. Combined scoring and candidate completion improve NDCG@10 across all three datasets with both T5 and decoder-only LC-Rec.

cs.IR↗

arXiv2026

RECAST: From Log Replay to Closed-Loop Driving Simulation with View-Complete Actors

Closed-loop driving simulation requires rendered observations to remain reliable as the ego vehicle and surrounding actors move beyond their recorded trajectories, exposing views absent from the source log. Existing data-driven simulators reconstruct dynamic actors from sparse observations, which can result in rendering artifacts under these viewpoint changes. We introduce RECAST (REconstructing Controllable Actors for Simulation and Testing), a 3D Gaussian Splatting framework that generates a view-complete actor from a single segmented vehicle observation in a driving log and registers the generated actor in the reconstructed scene. RECAST supports planner-in-the-loop rendering under controlled ego-actor interactions. To adapt an image-to-3D prior to real vehicles, we further introduce RECAR, a dataset of approximately 20K real vehicles with 600K background-free RGBA images spanning diverse vehicle colors and types. We use two-stage adaptation to improve vehicle generation from real driving-log observations. At the actor level, RECAST reduces $\mathrm{FD}_{\mathrm{incep}}$ from 9.788 to 7.992 relative to unadapted TRELLIS. At the scene level, under actor motion beyond logged trajectories, RECAST reduces $\mathrm{FD}_{\mathrm{incep}}$ from 129.35 to 112.10 and increases $\mathrm{CLIP}_{\mathrm{margin}}$ ($\times1000$) from 0.14 to 3.47 relative to Street Gaussians. We demonstrate planner-in-the-loop simulation with the image-conditioned planner GTRS-Dense. Compared with native Street Gaussians actors, RECAST increases the no-collision (NC) rate from 22.2% (12/54) to 63.0% (34/54) and the mean minimum predicted time-to-collision (TTC) from 0.798 s to 2.150 s. These experiments show that RECAST supports closed-loop planner evaluation under controlled ego-actor interactions beyond log replay. Visit our project page at https://zijunkr.github.io/RECAST/

cs.RO↗

arXiv2026

HALO: A Physics-Aware LLM Agent Framework for Nanophotonic Design

Language models have recently been applied to nanophotonic design, but it remains unclear whether they can reliably translate optical objectives into simulation-ready designs, execute electromagnetic analysis, and revise decisions from numerical feedback. We introduce HALO, a physics-aware framework that couples language-model planners with typed design specifications, electromagnetic simulation, diagnostic evaluation, and optional reuse of prior failure trajectories in an iterative design loop. We further introduce HALO-Bench, a 52-task benchmark spanning lab-derived, paper-derived, and open-ended nanophotonic design tasks under a shared evaluation protocol. We compare three planner configurations: a Fixed Structured Workflow, an Autonomous Structured Agent using the same simulation interface, and an Autonomous Coding Agent that directly writes and executes simulation code. The Fixed Structured Workflow is the most token-efficient and exhibits no observed code- or path-level failures, while autonomous coding can achieve higher task success with stronger models at the cost of additional operational failures. We also study reuse of prior failed trajectories. On targeted multi-round tasks, retrieved failure feedback reduces both iterations to first success and total token use. These results clarify the tradeoffs between explicit interfaces, autonomous execution, and reusable design experience in scientific agents.

physics.optics↗

arXiv2026

GateMem: Benchmarking Memory Governance in Multi-Principal Shared-Memory Agents

Memory benchmarks for LLM agents largely assume single-user settings, leaving shared assistants for hospitals, workplaces, campuses, and households understudied. In these deployments, multiple principals write to a common memory pool and query it under different roles, scopes, and relationships, so memory quality requires governance as well as recall. We introduce GateMem, a benchmark for multi-principal shared-memory agents. GateMem jointly evaluates utility for legitimate long-horizon requests with state updates, access control across contextual authorization boundaries, and agent-facing active forgetting after explicit deletion requests. It spans medical, office, education, and household domains, with long-form multi-party episodes, incremental memory injection, hidden checkpoints, structured judging, and leak-target annotations. Across diverse baselines and backbone models, no method simultaneously achieves strong utility, robust access control, and reliable forgetting. Long-context prompting often yields the best governance score at high token cost, while retrieval-based and external-memory methods reduce cost yet still leak unauthorized or deleted information. These results show current memory agents remain far from reliable shared institutional deployment.

cs.LG↗

arXiv2026

Witnessing Spin-Orbital Entanglement using Resonant Inelastic X-Ray Scattering

Entanglement plays a central role in quantum technologies, yet its characterization and control in materials remain challenging. Recent developments in spectrum-based entanglement witnesses have enabled new strategies for quantifying many-body entanglement in macroscopic materials. Here, we develop a protocol for detecting spin-orbital entanglement using experiment-accessible resonant inelastic x-ray scattering (RIXS). Central to our approach is the construction of a Hermitian generator from experimentally measurable spectra, which allows us to compute the quantum Fisher information (QFI) available in spin--orbital systems. The resulting QFI provides upper bounds for $k$-producible states and thus serves as a robust witness of spin-orbital entanglement. To account for realistic experimental limitations, we further extend our framework to include relaxed QFI bounds applicable to measurements lacking full polarization resolution.

quant-ph↗

arXiv2026

One-Step Relativistic Driven Similarity Renormalization Group Multireference Perturbation Theory

We present an efficient implementation of a one-step relativistic second-order multireference perturbation theory based on the multireference driven similarity renormalization group (MR-DSRG) using the exact two-component (X2C) Hamiltonian, which we denote X2C-DSRG-MRPT2. We show that the X2C-DSRG-MRPT2 method can accurately capture spin--orbit coupling (SOC) effects in the electronic structure of strongly correlated systems containing elements across the periodic table. We further demonstrate that the X2C-DSRG-MRPT2 method, through its variational treatment of SOC effects, can yield spin--orbit splittings with mean absolute percentage errors consistently below 7% with respect to experimental values for systems containing up to sixth row elements. With its modest computational scaling (fifth power in system size) and high accuracy, X2C-DSRG-MRPT2 provides a promising avenue for the routine treatment of relativistic effects in strongly correlated molecular systems.

physics.chem-ph↗

arXiv2026

The Python Simulations of Chemistry Framework: 10 years of an open-source quantum chemistry project

Over the past decade, the Python-based Simulations of Chemistry Framework (PySCF) has developed into a widely used open-source platform for electronic structure theory and quantum chemical method development. This article reviews the major advances since the previous overview in 2020, covering new modules and methodology, infrastructure changes, and performance benchmarks.

physics.chem-ph↗

arXiv2025

Multireference equation-of-motion driven similarity renormalization group for X-ray photoelectron spectra

We formulate and implement the core-valence separated multireference equation-of-motion driven similarity renormalization group method (CVS-IP-EOM-DSRG) for simulating X-ray photoelectron spectra (XPS) of strongly correlated molecular systems. This method is numerically robust and computationally efficient, delivering accurate core-ionization energies with O(N^4) scaling relative to basis set size N in the EOM step. To ensure rigorous core intensivity, we propose a simple modification of the ground-state MR-DSRG formalism. We develop and compare three variants of the theory based on different approximations of the effective Hamiltonian: two derived from low-order perturbative methods (DSRG-MRPT2 and DSRG-MRPT3), and one from a non-perturbative scheme truncated to 1- and 2-body operators [MR-LDSRG(2)]. We benchmark the CVS-IP-EOM-DSRG methods by computing vertical core-ionization energies for a representative molecular test set and comparing results against established single-reference and multireference methods. To demonstrate the applicability of CVS-IP-EOM-DSRG to strongly correlated systems, we compute the potential energy curves and vibrationally resolved XPS of N2 and CO and the XPS of ozone. Comparison with experimental data and other high-level theoretical results shows that all three CVS-IP-EOM-DSRG variants accurately predict vertical ionization energies, but only DSRG-MRPT3 and MR-LDSRG(2) levels of theory reliably capture the full dissociation behavior and reproduce the experimental vibrational structure.

physics.chem-ph↗

arXiv2025

Multireference equation-of-motion driven similarity renormalization group: theoretical foundations and applications to ionized states

We present a formulation and implementation of an equation-of-motion (EOM) extension of the multireference driven similarity renormalization group (MR-DSRG) formalism for ionization potentials (IP-EOM-DSRG). The IP-EOM-DSRG formalism results in a Hermitian generalized eigenvalue problem, delivering accurate ionization potentials for systems with strongly correlated ground and excited states. The EOM step scales as $O(N^5)$ with the basis set size $N$, allowing for efficient calculation of spectroscopic properties, such as transition energies and intensities. The IP-EOM-DSRG formalism is combined with three truncation schemes of the parent MR-DSRG theory: an iterative nonperturbative method with up to two-body excitations [MR-LDSRG(2)] and second- and third-order perturbative approximations [DSRG-MRPT2/3]. We benchmark these variants by computing 1) the vertical valence ionization potentials of a series of small molecules at both equilibrium and stretched geometries; 2) the spectroscopic constants of several low-lying electronic states of the OH, CN, N2+, and CO+ radicals; and 3) the binding curves of low-lying electronic states of the CN radical. A comparison with experimental data and theoretical results shows that all three IP-EOM-DSRG methods accurately reproduce the vertical ionization potentials and spectroscopic constants of these systems. Notably, the DSRG-MRPT3 and MR-LDSRG(2) versions outperform several state-of-the-art multireference methods of comparable or higher cost.

physics.chem-ph↗

arXiv2025

TANGLED: Generating 3D Hair Strands from Images with Arbitrary Styles and Viewpoints

Hairstyles are intricate and culturally significant with various geometries, textures, and structures. Existing text or image-guided generation methods fail to handle the richness and complexity of diverse styles. We present TANGLED, a novel approach for 3D hair strand generation that accommodates diverse image inputs across styles, viewpoints, and quantities of input views. TANGLED employs a three-step pipeline. First, our MultiHair Dataset provides 457 diverse hairstyles annotated with 74 attributes, emphasizing complex and culturally significant styles to improve model generalization. Second, we propose a diffusion framework conditioned on multi-view linearts that can capture topological cues (e.g., strand density and parting lines) while filtering out noise. By leveraging a latent diffusion model with cross-attention on lineart features, our method achieves flexible and robust 3D hair generation across diverse input conditions. Third, a parametric post-processing module enforces braid-specific constraints to maintain coherence in complex structures. This framework not only advances hairstyle realism and diversity but also enables culturally inclusive digital avatars and novel applications like sketch-based 3D strand editing for animation and augmented reality.

cs.CV↗

arXiv2025

Equation-of-motion internally contracted multireference unitary coupled-cluster theory

The accurate computation of excited states remains a challenge in electronic structure theory, especially for systems with a ground state that requires a multireference treatment. In this work, we introduce a novel equation-of-motion (EOM) extension of the internally contracted multireference unitary coupled-cluster framework (ic-MRUCC), termed EOM-ic-MRUCC. EOM-ic-MRUCC follows the transform-then-diagonalize approach, in analogy to its non-unitary counterpart [Datta and Nooijen, J. Chem. Phys. 137, 204107 (2012)]. By employing a projective approach to optimize the ground state, the method retains additive separability and proper scaling with system size. We show that excitation energies are size intensive if the EOM operator satisfies the "killer" and the projective conditions. Furthermore, we propose to represent changes in reference state upon electron excitation via projected many-body operators that span active orbitals and show that the EOM equations formulated in this way are invariant with respect to active orbital rotations. We test the EOM-ic-MRUCC method truncated to single and double excitations by computing the potential energy curves for several excited states of a BeH$_2$ model system, the HF molecule, and water undergoing symmetric dissociation. Across these systems, our method delivers accurate excitation energies and potential energy curves within 5 m$E_\mathrm{h}$ (ca. 0.14 eV) from full configuration interaction. We find that truncating the Baker-Campbell-Hausdorff series to four-fold commutators contributes negligible errors (on the order of $10^{-5}$ $E_\mathrm{h}$ or less), offering a practical route to highly accurate excited-state calculations with reduced computational overhead.

physics.chem-ph↗

arXiv2024

Instant Facial Gaussians Translator for Relightable and Interactable Facial Rendering

We propose GauFace, a novel Gaussian Splatting representation, tailored for efficient animation and rendering of physically-based facial assets. Leveraging strong geometric priors and constrained optimization, GauFace ensures a neat and structured Gaussian representation, delivering high fidelity and real-time facial interaction of 30fps@1440p on a Snapdragon 8 Gen 2 mobile platform. Then, we introduce TransGS, a diffusion transformer that instantly translates physically-based facial assets into the corresponding GauFace representations. Specifically, we adopt a patch-based pipeline to handle the vast number of Gaussians effectively. We also introduce a novel pixel-aligned sampling scheme with UV positional encoding to ensure the throughput and rendering quality of GauFace assets generated by our TransGS. Once trained, TransGS can instantly translate facial assets with lighting conditions to GauFace representation, With the rich conditioning modalities, it also enables editing and animation capabilities reminiscent of traditional CG pipelines. We conduct extensive evaluations and user studies, compared to traditional offline and online renderers, as well as recent neural rendering methods, which demonstrate the superior performance of our approach for facial asset rendering. We also showcase diverse immersive applications of facial assets using our TransGS approach and GauFace representation, across various platforms like PCs, phones and even VR headsets.

cs.GR↗

arXiv2024

Forte: A Suite of Advanced Multireference Quantum Chemistry Methods

Forte is an open-source library specialized in multireference electronic structure theories for molecular systems and the rapid prototyping of new methods. This paper gives an overview of the capabilities of Forte, its software architecture, and examples of applications enabled by the methods it implements.

physics.chem-ph↗

arXiv2024

Towards accurate spin-orbit splittings from relativistic multireference electronic structure theory

Most nonrelativistic electron correlation methods can be adapted to account for relativistic effects, as long as the relativistic molecular spinor integrals are available, from either a four-, two-, or one-component mean-field calculation. However, relativistic multireference correlation methods remain a relatively unexplored area, with mixed evidence regarding the improvements brought by perturbative treatments. We report, for the first time, the implementation of state-averaged four-component multireference perturbation theories to second and third order based on the driven similarity renormalization group (DSRG). With our methods, named 4c-SA-DSRG-MRPT2 and 3, we find that the dynamical correlation included on top of 4c-CASSCF references can significantly improve the spin-orbit splittings in p-block elements and potential energy surfaces when compared to 4c-CASSCF and 4c-CASPT2 results. We further show that 4c-DSRG-MRPT2 and 3 are applicable to these systems over a wide range of the flow parameter, with systematic improvement from second to third order in terms of both improved error statistics and reduced sensitivity with respect to the flow parameter.

cond-mat.str-el↗