Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userEnabling Rapid Development of Parallel Tree Search Applications
190
0
0.0
(
0
)
Ask ChatGPT about the research
No Arabic abstract
Virtual observatories will give astronomers easy access to an unprecedented amount of data. Extracting scientific knowledge from these data will increasingly demand both efficient algorithms as well as the power of parallel computers. Nearly all efficient analyses of large astronomical datasets use trees as their fundamental data structure. Writing efficient tree-based techniques, a task that is time-consuming even on single-processor computers, is exceedingly cumbersome on massively parallel platforms (MPPs). Most applications that run on MPPs are simulation codes, since the expense of developing them is offset by the fact that they will be used for many years by many researchers. In contrast, data analysis codes change far more rapidly, are often unique to individual researchers, and therefore accommodate little reuse. Consequently, the economics of the current high-performance computing development paradigm for MPPs does not favor data analysis applications. We have therefore built a library, called Ntropy, that provides a flexible, extensible, and easy-to-use way of developing tree-based data analysis algorithms for both serial and parallel platforms. Our experience has shown that not only does our library save development time, it can also deliver excellent serial performance and parallel scalability. Furthermore, Ntropy makes it easy for an astronomer with little or no parallel programming experience to quickly scale their application to a distributed multiprocessor environment. By minimizing development time for efficient and scalable data analysis, we enable wide-scale knowledge discovery on massive datasets.
rate research
Read More
Formal verification of neural networks (NNs) is a challenging and important problem. Existing efficient complete solvers typically require the branch-and-bound (BaB) process, which splits the problem domain into sub-domains and solves each sub-domain using faster but weaker incomplete verifiers, such as Linear Programming (LP) on linearly relaxed sub-domains. In this paper, we propose to use the backward mode linear relaxation based perturbation analysis (LiRPA) to replace LP during the BaB process, which can be efficiently implemented on the typical machine learning accelerators such as GPUs and TPUs. However, unlike LP, LiRPA when applied naively can produce much weaker bounds and even cannot check certain conflicts of sub-domains during splitting, making the entire procedure incomplete after BaB. To address these challenges, we apply a fast gradient based bound tightening procedure combined with batch splits and the design of minimal usage of LP bound procedure, enabling us to effectively use LiRPA on the accelerator hardware for the challenging complete NN verification problem and significantly outperform LP-based approaches. On a single GPU, we demonstrate an order of magnitude speedup compared to existing LP-based approaches.
Tree Search (TS) is crucial to some of the most influential successes in reinforcement learning. Here, we tackle two major challenges with TS that limit its usability: textit{distribution shift} and textit{scalability}. We first discover and analyze a counter-intuitive phenomenon: action selection through TS and a pre-trained value function often leads to lower performance compared to the original pre-trained agent, even when having access to the exact state and reward in future steps. We show this is due to a distribution shift to areas where value estimates are highly inaccurate and analyze this effect using Extreme Value theory. To overcome this problem, we introduce a novel off-policy correction term that accounts for the mismatch between the pre-trained value and its corresponding TS policy by penalizing under-sampled trajectories. We prove that our correction eliminates the above mismatch and bound the probability of sub-optimal action selection. Our correction significantly improves pre-trained Rainbow agents without any further training, often more than doubling their scores on Atari games. Next, we address the scalability issue given by the computational complexity of exhaustive TS that scales exponentially with the tree depth. We introduce Batch-BFS: a GPU breadth-first search that advances all nodes in each depth of the tree simultaneously. Batch-BFS reduces runtime by two orders of magnitude and, beyond inference, enables also training with TS of depths that were not feasible before. We train DQN agents from scratch using TS and show improvement in several Atari games compared to both the original DQN and the more advanced Rainbow.
While the open-source software development model has led to successful large-scale collaborations in building software systems, data science projects are frequently developed by individuals or small teams. We describe challenges to scaling data science collaborations and present a conceptual framework and ML programming model to address them. We instantiate these ideas in Ballet, a lightweight framework for collaborative, open-source data science through a focus on feature engineering, and an accompanying cloud-based development environment. Using our framework, collaborators incrementally propose feature definitions to a repository which are each subjected to an ML performance evaluation and can be automatically merged into an executable feature engineering pipeline. We leverage Ballet to conduct a case study analysis of an income prediction problem with 27 collaborators, and discuss implications for future designers of collaborative projects.
The current bulk materials discovery cycle has several inefficiencies from initial computational predictions through fabrication and analyses. Materials are generally evaluated in a singular fashion, relying largely on human-driven compositional choices and analysis of the volumes of generated data, thus also slowing validation of computational models. To overcome these limitations, we developed a high-throughput rapid experimental alloy development (HT-READ) methodology that comprises an integrated, closed-loop material screening process inspired by broad chemical assays and modern innovations in automation. Our method is a general framework unifying computational identification of ideal candidate materials, fabrication of sample libraries in a configuration amenable to multiple tests and processing routes, and analysis of the candidate materials in a high-throughput fashion. An artificial intelligence agent is used to find connections between compositions and material properties. New experimental data can be leveraged in subsequent iterations or new design objectives. The sample libraries are assigned unique identifiers and stored to make data and samples persistent, thus preventing institutional knowledge loss.
In this paper we present t
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
