arXiv subjects

Cunxi Yu

Publications and source records attributed to Cunxi Yu.

At least 55 records · Page 3Linked to original sources

arXiv2023

Accelerating Exact Combinatorial Optimization via RL-based Initialization -- A Case Study in Scheduling

Scheduling on dataflow graphs (also known as computation graphs) is an NP-hard problem. The traditional exact methods are limited by runtime complexity, while reinforcement learning (RL) and heuristic-based approaches struggle with determinism and solution quality. This research aims to develop an innovative approach that employs machine learning (ML) for addressing combinatorial optimization problems, using scheduling as a case study. The goal is to provide guarantees in optimality and determinism while maintaining the runtime cost of heuristic methods. Specifically, we introduce a novel two-phase RL-to-ILP scheduling framework, which includes three steps: 1) RL solver acts as coarse-grain scheduler, 2) solution relaxation and 3) exact solving via ILP. Our framework demonstrates the same scheduling performance compared with using exact scheduling methods while achieving up to 128 $\times$ speed improvements. This was conducted on actual EdgeTPU platforms, utilizing ImageNet DNN computation graphs as input. Additionally, the framework offers improved on-chip inference runtime and acceleration compared to the commercially available EdgeTPU compiler.

cs.LG↗

arXiv2023

Gamora: Graph Learning based Symbolic Reasoning for Large-Scale Boolean Networks

Reasoning high-level abstractions from bit-blasted Boolean networks (BNs) such as gate-level netlists can significantly benefit functional verification, logic minimization, datapath synthesis, malicious logic identification, etc. Mostly, conventional reasoning approaches leverage structural hashing and functional propagation, suffering from limited scalability and inefficient usage of modern computing power. In response, we propose a novel symbolic reasoning framework exploiting graph neural networks (GNNs) and GPU acceleration to reason high-level functional blocks from gate-level netlists, namely Gamora, which offers high reasoning performance w.r.t exact reasoning algorithms, strong scalability to BNs with over 33 million nodes, and generalization capability from simple to complex designs. To further demonstrate the capability of Gamora, we also evaluate its reasoning performance after various technology mapping options, since technology-dependent optimizations are known to make functional reasoning much more challenging. Experimental results show that (1) Gamora reaches almost 100% and over 97% reasoning accuracy for carry-save-array (CSA) and Booth-encoded multipliers, respectively, with up to six orders of magnitude speedups compared to the state-of-the-art implementation in the ABC framework; (2) Gamora maintains high reasoning accuracy (>92%) in finding functional modules after complex technology mapping, upon which we comprehensively analyze the impacts on Gamora reasoning from technology mapping.

cs.AR↗

arXiv2023

FlowTune: End-to-end Automatic Logic Optimization Exploration via Domain-specific Multi-armed Bandit

Recent years have seen increasing employment of decision intelligence in electronic design automation (EDA), which aims to reduce the manual efforts and boost the design closure process in modern toolflows. However, existing approaches either require a large number of labeled data and expensive training efforts, or are limited in practical EDA toolflow integration due to computation overhead. This paper presents a generic end-to-end sequential decision making framework FlowTune for synthesis tooflow optimization, with a novel high-performance domain-specific, multi-stage multi-armed bandit (MAB) approach. This framework addresses optimization problems on Boolean optimization problems such as a) And-Inv-Graphs (# nodes), b) Conjunction Normal Form (CNF) minimization (# clauses) for Boolean Satisfiability; logic synthesis and technology mapping problems such as c) post static timing analysis (STA) delay and area optimization for standard-cell technology mapping, and d) FPGA technology mapping for 6-in LUT architectures. Moreover, we demonstrate the high extnsibility and generalizability of the proposed domain-specific MAB approach with end-to-end FPGA design flow, evaluated at post-routing stage, with two different FPGA backend tools (OpenFPGA and VPR) and two different logic synthesis representations (AIGs and MIGs). FlowTune is fully integrated with ABC [1], Yosys [2], VTR [3], LSOracle [4], OpenFPGA [5], and industrial tools, and is released publicly. The experimental results conducted on various design stages in the flow all demonstrate that our framework outperforms both hand-crafted flows [1] and ML explored flows [6], [7] in quality of results, and is orders of magnitude faster compared to ML-based approaches.

cs.AR↗

arXiv2023

Rubik's Optical Neural Networks: Multi-task Learning with Physics-aware Rotation Architecture

Recently, there are increasing efforts on advancing optical neural networks (ONNs), which bring significant advantages for machine learning (ML) in terms of power efficiency, parallelism, and computational speed. With the considerable benefits in computation speed and energy efficiency, there are significant interests in leveraging ONNs into medical sensing, security screening, drug detection, and autonomous driving. However, due to the challenge of implementing reconfigurability, deploying multi-task learning (MTL) algorithms on ONNs requires re-building and duplicating the physical diffractive systems, which significantly degrades the energy and cost efficiency in practical application scenarios. This work presents a novel ONNs architecture, namely, \textit{RubikONNs}, which utilizes the physical properties of optical systems to encode multiple feed-forward functions by physically rotating the hardware similarly to rotating a \textit{Rubik's Cube}. To optimize MTL performance on RubikONNs, two domain-specific physics-aware training algorithms \textit{RotAgg} and \textit{RotSeq} are proposed. Our experimental results demonstrate more than 4$\times$ improvements in energy and cost efficiency with marginal accuracy degradation compared to the state-of-the-art approaches.

cs.LG↗

arXiv2023

RESPECT: Reinforcement Learning based Edge Scheduling on Pipelined Coral Edge TPUs

Deep neural networks (DNNs) have substantial computational and memory requirements, and the compilation of its computational graphs has a great impact on the performance of resource-constrained (e.g., computation, I/O, and memory-bound) edge computing systems. While efficient execution of their computational graph requires an effective scheduling algorithm, generating the optimal scheduling solution is a challenging NP-hard problem. Furthermore, the complexity of scheduling DNN computational graphs will further increase on pipelined multi-core systems considering memory communication cost, as well as the increasing size of DNNs. Using the synthetic graph for the training dataset, this work presents a reinforcement learning (RL) based scheduling framework RESPECT, which learns the behaviors of optimal optimization algorithms and generates near-optimal scheduling results with short solving runtime overhead. Our framework has demonstrated up to $\sim2.5\times$ real-world on-chip inference runtime speedups over the commercial compiler with ten popular ImageNet models deployed on the physical Coral Edge TPUs system. Moreover, compared to the exact optimization methods, the proposed RL scheduling improves the scheduling optimization runtime by up to 683$\times$ speedups compared to the commercial compiler and matches the exact optimal solutions with up to 930$\times$ speedups. Finally, we perform a comprehensive generalizability test, which demonstrates RESPECT successfully imitates optimal solving behaviors from small synthetic graphs to large real-world DNNs computational graphs.

cs.AR↗

arXiv2023

Physics-aware Roughness Optimization for Diffractive Optical Neural Networks

As a representative next-generation device/circuit technology beyond CMOS, diffractive optical neural networks (DONNs) have shown promising advantages over conventional deep neural networks due to extreme fast computation speed (light speed) and low energy consumption. However, there is a mismatch, i.e., significant prediction accuracy loss, between the DONN numerical modelling and physical optical device deployment, because of the interpixel interaction within the diffractive layers. In this work, we propose a physics-aware diffractive optical neural network training framework to reduce the performance difference between numerical modeling and practical deployment. Specifically, we propose the roughness modeling regularization in the training process and integrate the physics-aware sparsification method to introduce sparsity to the phase masks to reduce sharp phase changes between adjacent pixels in diffractive layers. We further develop $2π$ periodic optimization to reduce the roughness of the phase masks to preserve the performance of DONN. Experiment results demonstrate that, compared to state-of-the-arts, our physics-aware optimization can provide $35.7\%$, $34.2\%$, $28.1\%$, and $27.3\%$ reduction in roughness with only accuracy loss on MNIST, FMNIST, KMNIST, and EMNIST, respectively.

cs.LG↗

arXiv2022

Physics-aware Differentiable Discrete Codesign for Diffractive Optical Neural Networks

Diffractive optical neural networks (DONNs) have attracted lots of attention as they bring significant advantages in terms of power efficiency, parallelism, and computational speed compared with conventional deep neural networks (DNNs), which have intrinsic limitations when implemented on digital platforms. However, inversely mapping algorithm-trained physical model parameters onto real-world optical devices with discrete values is a non-trivial task as existing optical devices have non-unified discrete levels and non-monotonic properties. This work proposes a novel device-to-system hardware-software codesign framework, which enables efficient physics-aware training of DONNs w.r.t arbitrary experimental measured optical devices across layers. Specifically, Gumbel-Softmax is employed to enable differentiable discrete mapping from real-world device parameters into the forward function of DONNs, where the physical parameters in DONNs can be trained by simply minimizing the loss function of the ML task. The results have demonstrated that our proposed framework offers significant advantages over conventional quantization-based methods, especially with low-precision optical devices. Finally, the proposed algorithm is fully verified with physical experimental optical systems in low-precision settings.

cs.AI↗

arXiv2022

Physics-aware Complex-valued Adversarial Machine Learning in Reconfigurable Diffractive All-optical Neural Network

Diffractive optical neural networks have shown promising advantages over electronic circuits for accelerating modern machine learning (ML) algorithms. However, it is challenging to achieve fully programmable all-optical implementation and rapid hardware deployment. Furthermore, understanding the threat of adversarial ML in such system becomes crucial for real-world applications, which remains unexplored. Here, we demonstrate a large-scale, cost-effective, complex-valued, and reconfigurable diffractive all-optical neural networks system in the visible range based on cascaded transmissive twisted nematic liquid crystal spatial light modulators. With the assist of categorical reparameterization, we create a physics-aware training framework for the fast and accurate deployment of computer-trained models onto optical hardware. Furthermore, we theoretically analyze and experimentally demonstrate physics-aware adversarial attacks onto the system, which are generated from a complex-valued gradient-based algorithm. The detailed adversarial robustness comparison with conventional multiple layer perceptrons and convolutional neural networks features a distinct statistical adversarial property in diffractive optical neural networks. Our full stack of software and hardware provides new opportunities of employing diffractive optics in a variety of ML tasks and enabling the research on optical adversarial ML.

cs.ET↗

arXiv2022

Device-system Co-design of Photonic Neuromorphic Processor using Reinforcement Learning

The incorporation of high-performance optoelectronic devices into photonic neuromorphic processors can substantially accelerate computationally intensive operations in machine learning (ML) algorithms. However, the conventional device design wisdom is disconnected with system optimization. We report a device-system co-design methodology to optimize a free-space optical general matrix multiplication (GEMM) hardware accelerator by engineering a spatially reconfigurable array made from chalcogenide phase change materials. With a highly-parallelized hardware emulator constructed based on experimental information, we demonstrate the design of unit device by optimizing GEMM calculation accuracy via reinforcement learning, including deep Q-learning neural network, Bayesian optimization, and their cascaded approach, which show a clear correlation between system performance metrics and physical device specifications. Furthermore, we employ physics-aware training approaches to deploy optimized hardware to the tasks of image classification, materials discovery, and a closed-loop design of optical ML accelerators. The demonstrated framework offers insights into the co-design of optoelectronic devices and systems with reduced human-supervision and domain-knowledge barriers.

cs.ET↗

arXiv2021

Physics-Guided and Physics-Explainable Recurrent Neural Network for Time Dynamics in Optical Resonances

Understanding the time evolution of physical systems is crucial to revealing fundamental characteristics that are hidden in frequency domain. In optical science, high-quality resonance cavities and enhanced interactions with matters are at the heart of modern quantum technologies. However, capturing their time dynamics in real-world scenarios suffers from long data acquisition and low analysis accuracy due to slow convergence and limited time window. Here, we report a physics-guided and physics-explainable recurrent neural network to precisely forecast the time-domain response of resonance features with the shortest acquired input sequence being 7\% of full length, and to infer corresponding resonance frequencies. The model is trained in a two-step multi-fidelity framework for high-accuracy forecast, where the first step is based on a large amount of low-fidelity physical-model-generated synthetic data and second step involves a small set of high-fidelity application-oriented observational data. Through both simulations and experiments, we demonstrate that the model is universally applicable to a wide range of resonances, including dielectric metasurfaces, graphene plasmonics, and ultrastrongly coupled Landau polaritons, where our model accurately captures small signal features and learns essential physical quantities. The demonstrated machine learning algorithm offers a new way to accelerate the exploration of physical phenomena and the design of devices under resonance-enhanced light-matter interaction.

physics.optics↗

arXiv2021

Real-time Multi-Task Diffractive Deep Neural Networks via Hardware-Software Co-design

Deep neural networks (DNNs) have substantial computational requirements, which greatly limit their performance in resource-constrained environments. Recently, there are increasing efforts on optical neural networks and optical computing based DNNs hardware, which bring significant advantages for deep learning systems in terms of their power efficiency, parallelism and computational speed. Among them, free-space diffractive deep neural networks (D$^2$NNs) based on the light diffraction, feature millions of neurons in each layer interconnected with neurons in neighboring layers. However, due to the challenge of implementing reconfigurability, deploying different DNNs algorithms requires re-building and duplicating the physical diffractive systems, which significantly degrades the hardware efficiency in practical application scenarios. Thus, this work proposes a novel hardware-software co-design method that enables robust and noise-resilient Multi-task Learning in D$^2$NNs. Our experimental results demonstrate significant improvements in versatility and hardware efficiency, and also demonstrate the robustness of proposed multi-task D$^2$NN architecture under wide noise ranges of all system components. In addition, we propose a domain-specific regularization algorithm for training the proposed multi-task architecture, which can be used to flexibly adjust the desired performance for each task.

cs.LG↗

arXiv2020

Artificial Intelligence Accelerators based on Graphene Optoelectronic Devices

Optical and optoelectronic approaches of performing matrix-vector multiplication (MVM) operations have shown the great promise of accelerating machine learning (ML) algorithms with unprecedented performance. The incorporation of nanomaterials into the system can further improve the performance thanks to their extraordinary properties, but the non-uniformity and variation of nanostructures in the macroscopic scale pose severe limitations for large-scale hardware deployment. Here, we report a new optoelectronic architecture consisting of spatial light modulators and photodetector arrays made from graphene to perform MVM. The ultrahigh carrier mobility of graphene, nearly-zero-power-consumption electro-optic control, and extreme parallelism suggest ultrahigh data throughput and ultralow-power consumption. Moreover, we develop a methodology of performing accurate calculations with imperfect components, laying the foundation for scalable systems. Finally, we perform a few representative ML algorithms, including singular value decomposition, support vector machine, and deep neural networks, to show the versatility and generality of our platform.

cs.ET↗

arXiv2020

Contrastive Weight Regularization for Large Minibatch SGD

The minibatch stochastic gradient descent method (SGD) is widely applied in deep learning due to its efficiency and scalability that enable training deep networks with a large volume of data. Particularly in the distributed setting, SGD is usually applied with large batch size. However, as opposed to small-batch SGD, neural network models trained with large-batch SGD can hardly generalize well, i.e., the validation accuracy is low. In this work, we introduce a novel regularization technique, namely distinctive regularization (DReg), which replicates a certain layer of the deep network and encourages the parameters of both layers to be diverse. The DReg technique introduces very little computation overhead. Moreover, we empirically show that optimizing the neural network with DReg using large-batch SGD achieves a significant boost in the convergence and improved generalization performance. We also demonstrate that DReg can boost the convergence of large-batch SGD with momentum. We believe that DReg can be used as a simple regularization trick to accelerate large-batch training in deep learning.

cs.LG↗

arXiv2019

Painting on Placement: Forecasting Routing Congestion using Conditional Generative Adversarial Nets

Physical design process commonly consumes hours to days for large designs, and routing is known as the most critical step. Demands for accurate routing quality prediction raise to a new level to accelerate hardware innovation with advanced technology nodes. This work presents an approach that forecasts the density of all routing channels over the entire floorplan, with features collected up to placement, using conditional GANs. Specifically, forecasting the routing congestion is constructed as an image translation (colorization) problem. The proposed approach is applied to a) placement exploration for minimum congestion, b) constrained placement exploration and c) forecasting congestion in real-time during incremental placement, using eight designs targeting a fixed FPGA architecture.

cs.LG↗

arXiv2019

Efficient Parallel Verification of Galois Field Multipliers

Galois field (GF) arithmetic is used to implement critical arithmetic components in communication and security-related hardware, and verification of such components is of prime importance. Current techniques for formally verifying such components are based on computer algebra methods that proved successful in verification of integer arithmetic circuits. However, these methods are sequential in nature and do not offer any parallelism. This paper presents an algebraic functional verification technique of gate-level GF (2m ) multipliers, in which verification is performed in bit-parallel fashion. The method is based on extracting a unique polynomial in Galois field of each output bit independently. We demonstrate that this method is able to verify an n-bit GF multiplier in n threads. Experiments performed on pre- and post-synthesized Mastrovito and Montgomery multipliers show high efficiency up to 571 bits.

cs.SC↗

arXiv2019

Developing Synthesis Flows Without Human Knowledge

Design flows are the explicit combinations of design transformations, primarily involved in synthesis, placement and routing processes, to accomplish the design of Integrated Circuits (ICs) and System-on-Chip (SoC). Mostly, the flows are developed based on the knowledge of the experts. However, due to the large search space of design flows and the increasing design complexity, developing Intellectual Property (IP)-specific synthesis flows providing high Quality of Result (QoR) is extremely challenging. This work presents a fully autonomous framework that artificially produces design-specific synthesis flows without human guidance and baseline flows, using Convolutional Neural Network (CNN). The demonstrations are made by successfully designing logic synthesis flows of three large scaled designs.

cs.DC↗

arXiv2019

Spectral Approach to Verifying Non-linear Arithmetic Circuits

This paper presents a fast and effective computer algebraic method for analyzing and verifying non-linear integer arithmetic circuits using a novel algebraic spectral model. It introduces a concept of algebraic spectrum, a numerical form of polynomial expression; it uses the distribution of coefficients of the monomials to determine the type of arithmetic function under verification. In contrast to previous works, the proof of functional correctness is achieved by computing an algebraic spectrum combined with a local rewriting of word-level polynomials. The speedup is achieved by propagating coefficients through the circuit using And-Inverter Graph (AIG) datastructure. The effectiveness of the method is demonstrated with experiments including standard and Booth multipliers, and other synthesized non-linear arithmetic circuits up to 1024 bits containing over 12 million gates.

cs.SC↗

arXiv2018

Performance Estimation of Synthesis Flows cross Technologies using LSTMs and Transfer Learning

Due to the increasing complexity of Integrated Circuits (ICs) and System-on-Chip (SoC), developing high-quality synthesis flows within a short market time becomes more challenging. We propose a general approach that precisely estimates the Quality-of-Result (QoR), such as delay and area, of unseen synthesis flows for specific designs. The main idea is training a Recurrent Neural Network (RNN) regressor, where the flows are inputs and QoRs are ground truth. The RNN regressor is constructed with Long Short-Term Memory (LSTM) and fully-connected layers. This approach is demonstrated with 1.2 million data points collected using 14nm, 7nm regular-voltage (RVT), and 7nm low-voltage (LVT) FinFET technologies with twelve IC designs. The accuracy of predicting the QoRs (delay and area) within one technology is $\boldsymbol{\geq}$\textbf{98.0}\% over $\sim$240,000 test points. To enable accurate predictions cross different technologies and different IC designs, we propose a transfer-learning approach that utilizes the model pre-trained with 14nm datasets. Our transfer learning approach obtains estimation accuracy $\geq$96.3\% over $\sim$960,000 test points, using only 100 data points for training.

cs.LG↗