An official website of the United States government
We propose a neural particle level set (Neural PLS) method to accommodate tracking and evolving dynamic neural representations. At the heart of our approach is a set of oriented particles serving dual roles of interface trackers and sampling seeders. These dynamic particles are used to evolve the interface and construct neural representations on a multi-resolution grid-hash structure to hybridize coarse sparse distance fields and multi-scale feature encoding. Based on these parallel implementations and neural-network-friendly architectures, our neural particle level set method combines the computational merits on both ends of the traditional particle level sets and the modern implicit neural representations, in terms of feature representation and dynamic tracking. We demonstrate the efficacy of our approach by showcasing its performance surpassing traditional level-set methods in both benchmark tests and physical simulations.
more »
« less
Multi-level Partition of Unity on Differentiable Moving Particles
We introduce a differentiable moving particle representation based on the multi-level partition of unity (MPU) to represent dynamic implicit geometries. At the core of our representation are two groups of particles, named feature particles and sample particles, which can move in space and produce dynamic surfaces according to external velocity fields or optimization gradients. These two particle groups iteratively guide and correct each other by alternating their roles as inputs and outputs. Each feature particle carries a set of coefficients for a local quadratic patch. These particle patches are assembled with partition-of-unity weights to derive a continuous implicit global shape. Each sampling particle carries its position and orientation, serving as dense surface samples for optimization tasks. Based on these moving particles, we develop a fully differentiable framework to infer and evolve highly detailed implicit geometries, enhanced by a multi-level background grid for particle adaptivity, across different inverse tasks. We demonstrated the efficacy of our representation through various benchmark comparisons with state-of-the-art neural representations, achieving lower memory consumption, fewer training iterations, and orders of magnitude higher accuracy in handling topologically complex objects and dynamic tracking tasks.
more »
« less
- Award ID(s):
- 2433307
- PAR ID:
- 10604182
- Publisher / Repository:
- Association for Computing Machinery (ACM)
- Date Published:
- Journal Name:
- ACM Transactions on Graphics
- Volume:
- 43
- Issue:
- 6
- ISSN:
- 0730-0301
- Format(s):
- Medium: X Size: p. 1-21
- Size(s):
- p. 1-21
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
We propose an approach to mapping tissue and vascular systems without the use of contrast agents, based on moving and measuring magnetic particles. To this end, we consider a swarm of particles in a 1D or 2D grid that can be tracked and controlled by an external agent. Control inputs are applied uniformly so that each particle experiences the same applied forces. We present algorithms for three tasks: (1) Mapping, i.e., building a representation of the free and obstacle regions of the workspace; (2) Subset Coverage, i.e., ensuring that at least one particle reaches each of a set of desired locations; and (3) Coverage, i.e., ensuring that every free region on the map is visited by at least one particle. These tasks relate to a large body of previous work from robot navigation, both from theory and practice, which is based on individual control. We provide theoretical insights that have potential relevance for fast MRI scans with magnetically controlled contrast media. In particular, we develop a fundamentally new approach for searching for an object at an unknown distance D, where the search is subject to two different and independent cost parameters for moving and for measuring. We show that regardless of the relative cost of these two operations, there is a simple O(log D/log log D)-competitive strategy, which is the best possible. Also, we provide practically useful and computationally efficient strategies for higher-dimensional settings. These algorithms extend to any number of particles and show that additional particles tend to reduce the mean and the standard deviation of the time required for each task.more » « less
-
The retina and primary visual cortex (V1) both exhibit diverse neural populations sensitive to diverse visual features. Yet it remains unclear how neural populations in each area partition stimulus space to span these features. One possibility is that neural populations are organized into discrete groups of neurons, with each group signaling a particular constellation of features. Alternatively, neurons could be continuously distributed across feature-encoding space. To distinguish these possibilities, we presented a battery of visual stimuli to the mouse retina and V1 while measuring neural responses with multi-electrode arrays. Using machine learning approaches, we developed a manifold embedding technique that captures how neural populations partition feature space and how visual responses correlate with physiological and anatomical properties of individual neurons. We show that retinal populations discretely encode features, while V1 populations provide a more continuous representation. Applying the same analysis approach to convolutional neural networks that model visual processing, we demonstrate that they partition features much more similarly to the retina, indicating they are more like big retinas than little brains.more » « less
-
As AI continues to grow, modern applications are becoming more data- and compute-intensive, driving the development of specialized AI chips to meet these demands. One example is AMD's AI Engine (AIE), a dedicated hardware system that includes a 2D array of high-frequency very-long instruction words (VLIW) vector processors to provide high computational throughput and reconfigurability. However, AIE's specialized architecture presents tremendous challenges in programming and compiler optimization. Existing AIE programming frameworks lack a clean abstraction to represent multi-level parallelism in AIE; programmers have to figure out the parallelism within a kernel, manually do the partition, and assign sub-tasks to different AIE cores to exploit parallelism. These significantly lower the programming productivity. Furthermore, some AIE architectures include FPGAs to provide extra flexibility, but there is no unified intermediate representation (IR) that captures these architectural differences. As a result, existing compilers can only optimize the AIE portions of the code, overlooking potential FPGA bottlenecks and leading to suboptimal performance. To address these limitations, we introduce ARIES, an agile multi-level intermediate representation (MLIR) based compilation flow for reconfigurable devices with AIEs. ARIES introduces a novel programming model that allows users to map kernels to separate AIE cores, exploiting task- and tile-level parallelism without restructuring code. It also includes a declarative scheduling interface to explore instruction-level parallelism within each core. At the IR level, we propose a unified MLIR-based representation for AIE architectures, both with or without FPGA, facilitating holistic optimization and better portability across AIE device families. For the General Matrix Multiply (GEMM) benchmark, ARIES achieves 4.92 TFLOPS, 15.86 TOPS, and 45.94 TOPS throughput under FP32, INT16, and, INT8 data types on Versal VCK190 respectively. Compared with the state-of-the-art (SOTA) work CHARM for AIE, ARIES improves the throughput by 1.17x, 1.59x, and 1.47x correspondingly. For ResNet residual layer, ARIES achieves up to 22.58x speedup compared with optimized SOTA work Riallto on Ryzen-AI NPU. ARIES is open-sourced on GitHub: https://github.com/arc-research-lab/Aries.more » « less
-
Slickwater hydraulic fracturing is becoming a prevalent approach to economically recovering shale hydrocarbon. It is very important to understand the proppant’s transport behavior during slickwater hydraulic fracturing treatment for effective creation of a desired propped fracture geometry. The currently available models are either oversimplified or have been performed at limited length scales to avoid high computational requirements. Another limitation is that the currently available hydraulic fracturing simulators are developed using only single-sized proppant particles. Motivated by this, in this work, a computationally efficient, three-dimensional, multiphase particle-in-cell (MP-PIC) model was employed to simulate the multi-size proppant transport in a field-scale geometry using the Eulerian–Lagrangian framework. Instead of tracking each particle, groups of particles (called parcels) are tracked, which allows one to simulate the proppant transport in field-scale geometries at an affordable computational cost. Then, we found from our sensitivity study that pumping schedules significantly affect propped fracture surface area and average fracture conductivity, thereby influencing shale gas production. Motivated by these results, we propose an optimization framework using the MP-PIC model to design the multi-size proppant pumping schedule that maximizes shale gas production from unconventional reservoirs for given fracturing resources.more » « less
