arXiv subjects

Haoran Jin

Publications and source records attributed to Haoran Jin.

12 recordsLinked to original sources

arXiv2026

Fengshui: Demystifying Chiplet Ecosystem and Bespoke Neural Network Accelerator Codesign

Modern ML workloads, with stringent latency and energy constraints, are increasingly hard to run efficiently on homogeneous commodity hardware. We argue that operator-level disaggregation--tailoring microarchitecture, batching, and memory hierarchy to each operator--is essential to overcome these limitations, though the resulting highly bespoke accelerators incur prohibitive Non-Recurring Engineering (NRE) costs. Chiplet-based integration amortizes NRE across applications, but choosing which chiplets to build and how to compose them into accelerators is circularly dependent--a chiplet pool's value depends on the constructed accelerators, while accelerator quality is constrained by available chiplets. This paper introduces Fengshui, a chiplet ecosystem and accelerator co-design framework that jointly optimizes chiplet pool composition and bespoke application-specific integrated circuit (BASIC) design. Fengshui constructs BASICs through operator-level disaggregation, co-exploring chiplet and memory heterogeneity, tensor fusion, and pipeline/tensor/expert parallelism with place-and-route validation for physical implementability. With just 8 strategically selected chiplets, encompassing network switches, processing-in-memory units, and accelerators with diverse microarchitectures, Fengshui-generated BASICs achieve 48.5%, 88.1%, 93.0%, and 97.8% reductions in energy, energy-cost product (EC), energy-delay product (EDP), and energy-delay-cost product (EDPC) over homogeneous accelerators, while scoring within 4.1% of unconstrained heterogeneous designs across diverse neural networks. For datacenter MoE and dense LLM serving, Fengshui reduces prefill energy and EC by up to 16.8% and 28.7%, respectively; for edge autonomous vehicle perception, it achieves 12.0% energy and 23.6% EC reductions under real-time latency constraints.

cs.AR↗

arXiv2026

Rethinking Agentic Kernel Generation for Emerging Accelerators

Emerging accelerators often lack mature compiler backends, motivating neural agents that generate and repair kernels from architectural documentation and simulator feedback. This approach repeatedly reconstructs workload-invariant machine semantics--including instruction behavior, legality constraints, synchronization rules, and memory protocols--for every workload. We argue that these semantics should be compiled once into a persistent symbolic artifact, while neural reasoning should focus on workload-dependent mapping decisions. We present Zomboss, a compiler-mediated agentic framework for kernel generation that places neural search within a verified compiler boundary. Zomboss compiles machine semantics and legality constraints into a reusable mapping interface, then uses a neural agent to optimize workload-dependent decisions within the validated mapping space. Across 20 Gemmini and 36 PLENA workload instances, Zomboss returns a correct verified kernel on all 56 instances. Relative to the compiler default, Zomboss achieves geometric-mean speedups of $3.34\times$ on Gemmini and $1.10\times$ on PLENA. Relative to direct agentic generation, it reduces inference tokens by 71.2% on Gemmini and 54.2% on PLENA. These results show that a compiler-defined symbolic interface turns native kernel synthesis into verified design-space exploration: compiler infrastructure preserves legality and correctness, while neural guidance improves workload-specific performance with lower search cost and complete coverage.

cs.AR↗

arXiv2026

C$^{2}$R: Cross-sample Consistency Regularization Mitigates Feature Splitting and Absorption in Sparse Autoencoders

Sparse Autoencoders (SAEs) are widely used to interpret large language models by decomposing activations into sparse, human-understandable features, but scaling to large dictionaries exposes fundamental challenges. Systematic studies reveal pervasive feature splitting that fragments coherent concepts into non-atomic latents and widespread feature absorption that creates arbitrary exceptions in general features, severely compromising latent reliability. These issues stem from inconsistent latent assignment across samples: without cross-sample constraints, per-sample optimization often allows a single underlying concept to be inconsistently distributed across multiple redundant or interfering latents. To address this, we introduce C$^2$R (\underline{\textbf{C}}ross-sample \underline{\textbf{C}}onsistency \underline{\textbf{R}}egularization). C$^2$R explicitly encourages that each semantic feature is consistently represented by a unified latent across the batch by penalizing the co-activation of directionally similar latents. Comprehensive evaluation demonstrates that C$^2$R effectively mitigates both splitting and absorption while, crucially, preserving reconstruction fidelity, providing a principled solution that enhances latent interpretability without degrading model performance. Source code is available at https://github.com/hr-jin/Cross-sample-Consistency-Regularization.

cs.LG↗

arXiv2026

TensorLift: Automatic Extraction of Tensor-Level ISA Semantics from Accelerator RTL via MLIR Semantic Lifting

Most proposed tensor accelerators lack well-documented ISAs and compiler backends, and are exercised only through hand-written kernels covering a handful of operators. Recent work (TAIDL, ACT) shows that a tensor-level ISA specification is enough to generate complete software stacks automatically. Writing that specification, however, remains a manual, expert-driven process. We present TensorLift, the first end-to-end MLIR-based pipeline that lifts RTL-extracted accelerator semantics to TAIDL-like tensor ISA specifications. Building on prior architecture-level model extraction that yields bit-level IR, an 8-pass MLIR pipeline progressively recovers tensor structure (MAC idioms, saturation semantics, multi-dimensional buffer organizations, and layout transformations) and emits specifications the ACT ecosystem consumes directly. On Gemmini and VTA, TensorLift captures every hardware instruction the two designs decode across 156 MLIR files, reducing the extracted bit-level MLIR by 24.8% and 41.2% overall and by up to 92.9% on a processing element, with controller modules retaining irreducible control logic. It recovers hardware features the hand-written reference omits (multi-bank DMA configuration, pooling, and im2col) and lifts VTA unmodified. Core compute and data-movement semantics are proven equivalent to the RTL-extracted model by Z3 SMT, the remainder validated against golden simulator data. Fed into ACT, the extracted specification yields a compiler backend at parity with hand-written Gemmini kernels (1.014x geometric mean), giving an automated path from RTL to a working software stack.

cs.AR↗

arXiv2025

Mozart: A Chiplet Ecosystem-Accelerator Codesign Framework for Composable Bespoke Application Specific Integrated Circuits

Modern AI acceleration faces a fundamental challenge: conventional assumptions about memory requirements, batching effectiveness, and latency-throughput tradeoffs are systemwide generalizations that ignore the heterogeneous computational patterns of individual neural network operators. However, going towards network-level customization and operator-level heterogeneity incur substantial Non-Recurring Engineering (NRE) costs. While chiplet-based approaches have been proposed to amortize NRE costs, reuse opportunities remain limited without carefully identifying which chiplets are truly necessary. This paper introduces Mozart, a chiplet ecosystem and accelerator codesign framework that systematically constructs low cost bespoke application-specific integrated circuits (BASICs). BASICs leverage operator-level disaggregation to explore chiplet and memory heterogeneity, tensor fusion, and tensor parallelism, with place-and-route validation ensuring physical implementability. The framework also enables constraint-aware system-level optimization across deployment contexts ranging from datacenter inference serving to edge computing in autonomous vehicles. The evaluation confirms that with just 8 strategically selected chiplets, Mozart-generated composite BASICs achieve 43.5%, 25.4%, 67.7%, and 78.8% reductions in energy, energy-cost product, energy-delay product (EDP), and energy-delay-cost product compared to traditional homogeneous accelerators. For datacenter LLM serving, Mozart achieves 15-19% energy reduction and 35-39% energy-cost improvement. In speculative decoding, Mozart delivers throughput improvements of 24.6-58.6% while reducing energy consumption by 38.6-45.6%. For autonomous vehicle perception, Mozart reduces energy-cost by 25.54% and energy by 10.53% under real-time constraints.

cs.AR↗

arXiv2025

The 1st International Workshop on Disentangled Representation Learning for Controllable Generation (DRL4Real): Methods and Results

This paper reviews the 1st International Workshop on Disentangled Representation Learning for Controllable Generation (DRL4Real), held in conjunction with ICCV 2025. The workshop aimed to bridge the gap between the theoretical promise of Disentangled Representation Learning (DRL) and its application in realistic scenarios, moving beyond synthetic benchmarks. DRL4Real focused on evaluating DRL methods in practical applications such as controllable generation, exploring advancements in model robustness, interpretability, and generalization. The workshop accepted 9 papers covering a broad range of topics, including the integration of novel inductive biases (e.g., language), the application of diffusion models to DRL, 3D-aware disentanglement, and the expansion of DRL into specialized domains like autonomous driving and EEG analysis. This summary details the workshop's objectives, the themes of the accepted papers, and provides an overview of the methodologies proposed by the authors.

cs.LG↗

arXiv2025

Internal Value Alignment in Large Language Models through Controlled Value Vector Activation

Aligning Large Language Models (LLMs) with human values has attracted increasing attention since it provides clarity, transparency, and the ability to adapt to evolving scenarios. In this paper, we introduce a Controlled Value Vector Activation (ConVA) method that directly aligns the internal values of LLMs by interpreting how a value is encoded in their latent representations and modifies relevant activations to ensure consistent values in LLMs. To ensure an accurate and unbiased interpretation, we propose a context-controlled value vector identification method. To consistently control values without sacrificing model performance, we introduce a gated value vector activation method for effective and minimum degree of value control. Experiments show that our method achieves the highest control success rate across 10 basic values without hurting LLM performance and fluency, and ensures target values even with opposite and potentially malicious input prompts. Source code and data are available at~ https://github.com/hr-jin/ConVA.

cs.CL↗

arXiv2024

Deep Group Interest Modeling of Full Lifelong User Behaviors for CTR Prediction

Extracting users' interests from their lifelong behavior sequence is crucial for predicting Click-Through Rate (CTR). Most current methods employ a two-stage process for efficiency: they first select historical behaviors related to the candidate item and then deduce the user's interest from this narrowed-down behavior sub-sequence. This two-stage paradigm, though effective, leads to information loss. Solely using users' lifelong click behaviors doesn't provide a complete picture of their interests, leading to suboptimal performance. In our research, we introduce the Deep Group Interest Network (DGIN), an end-to-end method to model the user's entire behavior history. This includes all post-registration actions, such as clicks, cart additions, purchases, and more, providing a nuanced user understanding. We start by grouping the full range of behaviors using a relevant key (like item_id) to enhance efficiency. This process reduces the behavior length significantly, from O(10^4) to O(10^2). To mitigate the potential loss of information due to grouping, we incorporate two categories of group attributes. Within each group, we calculate statistical information on various heterogeneous behaviors (like behavior counts) and employ self-attention mechanisms to highlight unique behavior characteristics (like behavior type). Based on this reorganized behavior data, the user's interests are derived using the Transformer technique. Additionally, we identify a subset of behaviors that share the same item_id with the candidate item from the lifelong behavior sequence. The insights from this subset reveal the user's decision-making process related to the candidate item, improving prediction accuracy. Our comprehensive evaluation, both on industrial and public datasets, validates DGIN's efficacy and efficiency.

cs.IR↗

arXiv2024

Evaluating Readability and Faithfulness of Concept-based Explanations

With the growing popularity of general-purpose Large Language Models (LLMs), comes a need for more global explanations of model behaviors. Concept-based explanations arise as a promising avenue for explaining high-level patterns learned by LLMs. Yet their evaluation poses unique challenges, especially due to their non-local nature and high dimensional representation in a model's hidden space. Current methods approach concepts from different perspectives, lacking a unified formalization. This makes evaluating the core measures of concepts, namely faithfulness or readability, challenging. To bridge the gap, we introduce a formal definition of concepts generalizing to diverse concept-based explanations' settings. Based on this, we quantify the faithfulness of a concept explanation via perturbation. We ensure adequate perturbation in the high-dimensional space for different concepts via an optimization problem. Readability is approximated via an automatic and deterministic measure, quantifying the coherence of patterns that maximally activate a concept while aligning with human understanding. Finally, based on measurement theory, we apply a meta-evaluation method for evaluating these measures, generalizable to other types of explanations or tasks as well. Extensive experimental analysis has been conducted to inform the selection of explanation evaluation measures.

cs.AI↗

arXiv2024

Efficient Transfer Learning Framework for Cross-Domain Click-Through Rate Prediction

Natural content and advertisement coexist in industrial recommendation systems but differ in data distribution. Concretely, traffic related to the advertisement is considerably sparser compared to that of natural content, which motivates the development of transferring knowledge from the richer source natural content domain to the sparser advertising domain. The challenges include the inefficiencies arising from the management of extensive source data and the problem of 'catastrophic forgetting' that results from the CTR model's daily updating. To this end, we propose a novel tri-level asynchronous framework, i.e., Efficient Transfer Learning Framework for Cross-Domain Click-Through Rate Prediction (E-CDCTR), to transfer comprehensive knowledge of natural content to advertisement CTR models. This framework consists of three key components: Tiny Pre-training Model ((TPM), which trains a tiny CTR model with several basic features on long-term natural data; Complete Pre-training Model (CPM), which trains a CTR model holding network structure and input features the same as target advertisement on short-term natural data; Advertisement CTR model (A-CTR), which derives its parameter initialization from CPM together with multiple historical embeddings from TPM as extra feature and then fine-tunes on advertisement data. TPM provides richer representations of user and item for both the CPM and A-CTR, effectively alleviating the forgetting problem inherent in the daily updates. CPM further enhances the advertisement model by providing knowledgeable initialization, thereby alleviating the data sparsity challenges typically encountered by advertising CTR models. Such a tri-level cross-domain transfer learning framework offers an efficient solution to address both data sparsity and `catastrophic forgetting', yielding remarkable improvements.

cs.IR↗

arXiv2023

AT4CTR: Auxiliary Match Tasks for Enhancing Click-Through Rate Prediction

Click-through rate (CTR) prediction is a vital task in industrial recommendation systems. Most existing methods focus on the network architecture design of the CTR model for better accuracy and suffer from the data sparsity problem. Especially in industrial recommendation systems, the widely applied negative sample down-sampling technique due to resource limitation worsens the problem, resulting in a decline in performance. In this paper, we propose \textbf{A}uxiliary Match \textbf{T}asks for enhancing \textbf{C}lick-\textbf{T}hrough \textbf{R}ate prediction accuracy (AT4CTR) by alleviating the data sparsity problem. Specifically, we design two match tasks inspired by collaborative filtering to enhance the relevance modeling between user and item. As the "click" action is a strong signal which indicates the user's preference towards the item directly, we make the first match task aim at pulling closer the representation between the user and the item regarding the positive samples. Since the user's past click behaviors can also be treated as the user him/herself, we apply the next item prediction as the second match task. For both the match tasks, we choose the InfoNCE as their loss function. The two match tasks can provide meaningful training signals to speed up the model's convergence and alleviate the data sparsity. We conduct extensive experiments on one public dataset and one large-scale industrial recommendation dataset. The result demonstrates the effectiveness of the proposed auxiliary match tasks. AT4CTR has been deployed in the real industrial advertising system and has gained remarkable revenue.

cs.IR↗

arXiv2023

HSD-PAM: High Speed Super Resolution Deep Penetration Photoacoustic Microscopy Imaging Boosted by Dual Branch Fusion Network

Photoacoustic microscopy (PAM) is a novel implementation of photoacoustic imaging (PAI) for visualizing the 3D bio-structure, which is realized by raster scanning of the tissue. However, as three involved critical imaging parameters, imaging speed, lateral resolution, and penetration depth have mutual effect to one the other. The improvement of one parameter results in the degradation of other two parameters, which constrains the overall performance of the PAM system. Here, we propose to break these limitations by hardware and software co-design. Starting with low lateral resolution, low sampling rate AR-PAM imaging which possesses the deep penetration capability, we aim to enhance the lateral resolution and up sampling the images, so that high speed, super resolution, and deep penetration for the PAM system (HSD-PAM) can be achieved. Data-driven based algorithm is a promising approach to solve this issue, thereby a dedicated novel dual branch fusion network is proposed, which includes a high resolution branch and a high speed branch. Since the availability of switchable AR-OR-PAM imaging system, the corresponding low resolution, undersample AR-PAM and high resolution, full sampled OR-PAM image pairs are utilized for training the network. Extensive simulation and in vivo experiments have been conducted to validate the trained model, enhancement results have proved the proposed algorithm achieved the best perceptual and quantitative image quality. As a result, the imaging speed is increased 16 times and the imaging lateral resolution is improved 5 times, while the deep penetration merit of AR-PAM modality is still reserved.

eess.IV↗