-
Masked Autoencoders for Microscopy are Scalable Learners of Cellular Biology
Authors:
Oren Kraus,
Kian Kenyon-Dean,
Saber Saberian,
Maryam Fallah,
Peter McLean,
Jess Leung,
Vasudev Sharma,
Ayla Khan,
Jia Balakrishnan,
Safiye Celik,
Dominique Beaini,
Maciej Sypetkowski,
Chi Vicky Cheng,
Kristen Morse,
Maureen Makes,
Ben Mabey,
Berton Earnshaw
Abstract:
Featurizing microscopy images for use in biological research remains a significant challenge, especially for large-scale experiments spanning millions of images. This work explores the scaling properties of weakly supervised classifiers and self-supervised masked autoencoders (MAEs) when training with increasingly larger model backbones and microscopy datasets. Our results show that ViT-based MAEs…
▽ More
Featurizing microscopy images for use in biological research remains a significant challenge, especially for large-scale experiments spanning millions of images. This work explores the scaling properties of weakly supervised classifiers and self-supervised masked autoencoders (MAEs) when training with increasingly larger model backbones and microscopy datasets. Our results show that ViT-based MAEs outperform weakly supervised classifiers on a variety of tasks, achieving as much as a 11.5% relative improvement when recalling known biological relationships curated from public databases. Additionally, we develop a new channel-agnostic MAE architecture (CA-MAE) that allows for inputting images of different numbers and orders of channels at inference time. We demonstrate that CA-MAEs effectively generalize by inferring and evaluating on a microscopy image dataset (JUMP-CP) generated under different experimental conditions with a different channel structure than our pretraining data (RPI-93M). Our findings motivate continued research into scaling self-supervised learning on microscopy data in order to create powerful foundation models of cellular biology that have the potential to catalyze advancements in drug discovery and beyond.
△ Less
Submitted 15 April, 2024;
originally announced April 2024.
-
Masked Autoencoders are Scalable Learners of Cellular Morphology
Authors:
Oren Kraus,
Kian Kenyon-Dean,
Saber Saberian,
Maryam Fallah,
Peter McLean,
Jess Leung,
Vasudev Sharma,
Ayla Khan,
Jia Balakrishnan,
Safiye Celik,
Maciej Sypetkowski,
Chi Vicky Cheng,
Kristen Morse,
Maureen Makes,
Ben Mabey,
Berton Earnshaw
Abstract:
Inferring biological relationships from cellular phenotypes in high-content microscopy screens provides significant opportunity and challenge in biological research. Prior results have shown that deep vision models can capture biological signal better than hand-crafted features. This work explores how self-supervised deep learning approaches scale when training larger models on larger microscopy d…
▽ More
Inferring biological relationships from cellular phenotypes in high-content microscopy screens provides significant opportunity and challenge in biological research. Prior results have shown that deep vision models can capture biological signal better than hand-crafted features. This work explores how self-supervised deep learning approaches scale when training larger models on larger microscopy datasets. Our results show that both CNN- and ViT-based masked autoencoders significantly outperform weakly supervised baselines. At the high-end of our scale, a ViT-L/8 trained on over 3.5-billion unique crops sampled from 93-million microscopy images achieves relative improvements as high as 28% over our best weakly supervised baseline at inferring known biological relationships curated from public databases. Relevant code and select models released with this work can be found at: https://github.com/recursionpharma/maes_microscopy.
△ Less
Submitted 27 November, 2023; v1 submitted 27 September, 2023;
originally announced September 2023.
-
ERIC: An Efficient and Practical Software Obfuscation Framework
Authors:
Alperen Bolat,
Seyyid Hikmet Çelik,
Ataberk Olgun,
Oğuz Ergin,
Marco Ottavi
Abstract:
Modern cloud computing systems distribute software executables over a network to keep the software sources, which are typically compiled in a security-critical cluster, secret. We develop ERIC, a new, efficient, and general software obfuscation framework. ERIC protects software against (i) static analysis, by making only an encrypted version of software executables available to the human eye, no m…
▽ More
Modern cloud computing systems distribute software executables over a network to keep the software sources, which are typically compiled in a security-critical cluster, secret. We develop ERIC, a new, efficient, and general software obfuscation framework. ERIC protects software against (i) static analysis, by making only an encrypted version of software executables available to the human eye, no matter how the software is distributed, and (ii) dynamic analysis, by guaranteeing that an encrypted executable can only be correctly decrypted and executed by a single authenticated device. ERIC comprises key hardware and software components to provide efficient software obfuscation support: (i) a hardware decryption engine (HDE) enables efficient decryption of encrypted hardware in the target device, (ii) the compiler can seamlessly encrypt software executables given only a unique device identifier. Both the hardware and software components are ISA-independent, making ERIC general. The key idea of ERIC is to use physical unclonable functions (PUFs), unique device identifiers, as secret keys in encrypting software executables. Malicious parties that cannot access the PUF in the target device cannot perform static or dynamic analyses on the encrypted binary. We develop ERIC's prototype on an FPGA to evaluate it end-to-end. Our prototype extends RISC-V Rocket Chip with the hardware decryption engine (HDE) to minimize the overheads of software decryption. We augment the custom LLVM-based compiler to enable partial/full encryption of RISC-V executables. The HDE incurs minor FPGA resource overheads, it requires 2.63% more LUTs and 3.83% more flip-flops compared to the Rocket Chip baseline. LLVM-based software encryption increases compile time by 15.22% and the executable size by 1.59%. ERIC is publicly available and can be downloaded from https://github.com/kasirgalabs/ERIC
△ Less
Submitted 15 July, 2022;
originally announced July 2022.
