arXiv subjects

Sagar Patel

Publications and source records attributed to Sagar Patel.

7 recordsLinked to original sources

arXiv2023

CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems

We present CrystalBox, a novel, model-agnostic, posthoc explainability framework for Deep Reinforcement Learning (DRL) controllers in the large family of input-driven environments which includes computer systems. We combine the natural decomposability of reward functions in input-driven environments with the explanatory power of decomposed returns. We propose an efficient algorithm to generate future-based explanations across both discrete and continuous control environments. Using applications such as adaptive bitrate streaming and congestion control, we demonstrate CrystalBox's capability to generate high-fidelity explanations. We further illustrate its higher utility across three practical use cases: contrastive explanations, network observability, and guided reward design, as opposed to prior explainability techniques that identify salient features.

cs.LG↗

arXiv2023

Plume: A Framework for High Performance Deep RL Network Controllers via Prioritized Trace Sampling

Deep Reinforcement Learning (DRL) has shown promise in various networking environments. However, these environments present several fundamental challenges for standard DRL techniques. They are difficult to explore and exhibit high levels of noise and uncertainty. Although these challenges complicate the training process, we find that in practice we can substantially mitigate their effects and even achieve state-of-the-art real-world performance by addressing a factor that has been previously overlooked: the skewed input trace distribution in DRL training datasets. We introduce a generalized framework, Plume, to automatically identify and balance the skew using a three-stage process. First, we identify the critical features that determine the behavior of the traces. Second, we classify the traces into clusters. Finally, we prioritize the salient clusters to improve the overall performance of the controller. Plume seamlessly works across DRL algorithms, without requiring any changes to the DRL workflow. We evaluated Plume on three networking environments, including Adaptive Bitrate Streaming, Congestion Control, and Load Balancing. Plume offers superior performance in both simulation and real-world settings, across different controllers and DRL algorithms. For example, our novel ABR controller, Gelato trained with Plume consistently outperforms prior state-of-the-art controllers on the live streaming platform Puffer for over a year. It is the first controller on the platform to deliver statistically significant improvements in both video quality and stalling, decreasing stalls by as much as 75%.

cs.LG↗

arXiv2021

Feature Selection on a Flare Forecasting Testbed: A Comparative Study of 24 Methods

The Space-Weather ANalytics for Solar Flares (SWAN-SF) is a multivariate time series benchmark dataset recently created to serve the heliophysics community as a testbed for solar flare forecasting models. SWAN-SF contains 54 unique features, with 24 quantitative features computed from the photospheric magnetic field maps of active regions, describing their precedent flare activity. In this study, for the first time, we systematically attacked the problem of quantifying the relevance of these features to the ambitious task of flare forecasting. We implemented an end-to-end pipeline for preprocessing, feature selection, and evaluation phases. We incorporated 24 Feature Subset Selection (FSS) algorithms, including multivariate and univariate, supervised and unsupervised, wrappers and filters. We methodologically compared the results of different FSS algorithms, both on the multivariate time series and vectorized formats, and tested their correlation and reliability, to the extent possible, by using the selected features for flare forecasting on unseen data, in univariate and multivariate fashions. We concluded our investigation with a report of the best FSS methods in terms of their top-k features, and the analysis of the findings. We wish the reproducibility of our study and the availability of the data allow the future attempts be comparable with our findings and themselves.

astro-ph.SR↗

arXiv2021

Architectural bone parameters and the relationship to titanium lattice design for powder bed fusion additive manufacturing

Additive manufacturing (AM) of titanium (Ti) and Ti-6Al-4V lattices has been proposed for bone implants and augmentation devices. Ti and Ti-6Al-4V have favourable biocompatibility, corrosion resistance and fatigue strength for bone applications; yet, the optimal parameters for Ti-6Al-4V lattice designs corresponding to the natural micro- and meso-scale architecture of human trabecular and cortical bone are not well understood. A comprehensive review was completed to compare the natural lattice architecture properties in human bone to Ti and Ti-6Al-4V lattice structures for bone replacement and repair. Ti and Ti-6Al-4V lattice porosity has varied from 15% to 97% with most studies reporting a porosity between 50-70%. Cortical bone is roughly 5-15% porous and lattices with 50-70% porosity are able to achieve comparable stiffness, compressive strength, and yield strength. Trabecular bone has a reported porosity range from 70-90%, with trabecular thickness varying from 120-200 {\mu}m. Existing powder bed fusion technologies have produced strut and wall thicknesses ranging from 200-1669 {\mu}m. This suggests limited overlap between current AM of Ti and Ti-6Al-4V lattice structures and trabecular bone architecture, indicating that replicating natural trabecular bone parameters with latticing is prohibitively challenging. This review contributes to the body of knowledge by identifying the correspondence of Ti and Ti-6Al-4V lattices to the natural parameters of bone microarchitectures, and provides further guidance on the design and AM recommendations towards addressing recognized performance gaps with powder bed fusion technologies.

cond-mat.mtrl-sci↗

arXiv2021

The influence of beam focus during laser powder bed fusion of a high reflectivity aluminium alloy -- AlSi$_{10}$Mg

LPBF of Al alloys is associated with numerous challenges when compared to other commonly used alloys due to their higher reflectivity and thermal conductivity. In this work, processing diagrams, temperature prediction models, XCT, and metallography are used for establishing criteria in process parameter optimization of high reflectivity Al alloys based on AlSi$_{10}$Mg response in using 57 different process parameter combinations - 21 using a focused Gaussian laser beam and 36 using divergent beams. For LPBF systems with focused beam diameters <100 {\mu}m, divergent beams obtained by defocusing to a position above the LPBF build plate primarily lead to conduction mode melt pools, while a focused beam leads to transition and keyhole mode melt pools. Conduction mode melting helps in avoiding keyhole mode defects, resulting in parts with densities >99.98%. Additionally, an analytical model-guided selection of laser power and velocity settings for a focused beam help in stabilizing melt pool and spatter dynamics in the transition melting mode thereby enabling a potential to obtain density values close to conduction mode densities (~99.98%). A dimensionless keyhole number (Ke) was derived in this work to identify distinct regions of conduction (Ke of 0-12), transition (Ke of 12-20), and keyhole (Ke > 20) mode melting during LPBF of AlSi$_{10}$. A melt pool aspect ratio (ratio of melt pool depth to width) of ~0.4 is observed to be the threshold between conduction and transition/keyhole mode melt pools for AlSi$_{10}$, different from the conventionally assumed 0.5. Lastly, inferred laser absorptivity values (from experimental melt pools) of transition/keyhole mode melt pools are observed to be >40% higher when compared to conduction mode melt pools. This work demonstrates a dimensionless-process map method to obtain near fully dense parts that can be generalized for LPBF of high reflectivity alloys.

cond-mat.mtrl-sci↗

arXiv2020

Gene family amplification facilitates adaptation in freshwater Unionid bivalve Megalonaias nervosa

As organisms are faced with intense rapidly changing selective pressures, new genetic material is required to facilitate adaptation. Among sources of genetic novelty, gene duplications and transposable elements (TEs) offer new genes or new regulatory patterns that can facilitate evolutionary change. With advances in genome sequencing it is possible to gain a broader view of how gene family proliferation and TE content evolve in non-model species when populations become threatened. Freshwater bivalves (Unionidae) currently face severe anthropogenic challenges. Over 70% of species in the United States are threatened, endangered or extinct due to pollution, damming of waterways, and overfishing. We have created a reference genome for M. nervosa to determine how genome content has evolved in the face of these widespread environmental challenges. We observe a burst of recent transposable element proliferation causing a 382 Mb expansion in genome content. Gene family expansion is common, with a duplication rate of 1.16 x 10^-8 per gene per generation. Cytochrome P450, ABC transporters, Hsp70 genes, von Willebrand proteins, chitin metabolism genes, mitochondria eating proteins, and opsin gene families have experienced significantly greater amplification and show signatures of selection. We use evolutionary theory to assess the relative contribution of SNPs and duplications in evolutionary change. Estimates suggest that gene family evolution may offer an exceptional substrate of genetic variation in M. nervosa, with Psgv=0.185 compared with Psgv=0.067 for single nucleotide changes. Hence, we suggest that gene family evolution is a source of "hopeful monsters" within the genome that facilitate adaptation.

q-bio.GN↗

arXiv2018

3D Topology Optimization using Convolutional Neural Networks

Topology optimization is computationally demanding that requires the assembly and solution to a finite element problem for each material distribution hypothesis. As a complementary alternative to the traditional physics-based topology optimization, we explore a data-driven approach that can quickly generate accurate solutions. To this end, we propose a deep learning approach based on a 3D encoder-decoder Convolutional Neural Network architecture for accelerating 3D topology optimization and to determine the optimal computational strategy for its deployment. Analysis of iteration-wise progress of the Solid Isotropic Material with Penalization process is used as a guideline to study how the earlier steps of the conventional topology optimization can be used as input for our approach to predict the final optimized output structure directly from this input. We conduct a comparative study between multiple strategies for training the neural network and assess the effect of using various input combinations for the CNN to finalize the strategy with the highest accuracy in predictions for practical deployment. For the best performing network, we achieved about 40% reduction in overall computation time while also attaining structural accuracies in the order of 96%.

cs.LG↗