Maintaining software security in a project as the volume of new code increases is a pressing problem in big data. Once a vulnerability is identified in a piece of software, it may be essential to locate other similar weaknesses, but conducting this type of search is time consuming and challenging. Currently, tools used to aid in searches like this are only capable of finding exact matches, but similar vulnerabilities may be expressed using different semantics. To address this problem, we developed an approach to efficiently find code with similar functionality in large repositories of executables: Syntax-Agnostic Code Similarity (SACS). SACS…
NovettaNLP
A Powerful, Transformers-Based Framework for End-to-End Natural Language Processing Tasks
BERT’s release in November 2018 marked an inflection point for Natural Language Processing (NLP), moving us closer to human-level performance on a range of use cases. Since then, many transformer-based language models, such as XLM, GPT-2, XLNet, and roBERTa, have been released, with each subsequent model raising the performance ceiling. With the availability of these state-of-the-art pre-trained language models, we can produce NLP models that solve a variety of downstream language-based tasks, enabling our customers to quickly analyze large volumes of unstructured text. The challenge with using these models is that they each differ slightly in their application. For example,…
Adversarial Attacks in Machine Learning: Defenses
Part 3 of 3
Introduction to Adversarial Attack Defense Methods In our previous blog posts, we introduced and experimented with adversarial attacks against deep learning models. Here, we explore ways to mitigate or decrease the effectiveness of some of these attacks with three defense methods: 1. Modifying the feature space of the neural network itself 2. Modifying an input to a network before it is used for inference 3. Modifying the training data to account for adversarial inputs Modifying the Feature Space of a Neural Network In our first experiment, we implemented a technique [1] that modifies the feature space of the weights for…
Adversarial Attacks in Machine Learning: Experimentation
Part 2 of 3
As machine learning (ML) becomes more prevalent, bad actors seek to take advantage of our confidence in models’ predictions. An example of this can be found in autonomous vehicle safety, wherein the application of a small sticker to a stop sign causes the vehicles’ ML models to incorrectly classify and interpret the visual cues on the sign [1]. In an exercise to better understand common vulnerabilities in model deployment, we experimented with a handful of prevalent adversarial attacks to understand their viability and demonstrate their potential. Figure 1: A human can easily discern this as a stop sign with stickers…
Adversarial Attacks in Machine Learning: Invisible Threats
Part 1 of 3
Artificial intelligence (AI) and machine learning (ML) models have become ingrained in everyday life, with applications ranging from biometric verification to virtual assistants. Reliance on ML becomes problematic when models misclassify an observation, potentially leading to adverse consequences. Thus the performance, robustness, and security of these models is critical to their success. Modern deep learning models have surpassed human performance in some specific tasks (e.g. types of computer vision), though they can be highly vulnerable to adversarial attacks. Such attacks are often crafted using methods similar to those employed by the models that perform the intended classifications. For example, an…
ADSynth Blog Series: AWS Pipeline and iOS App Creation
Part 4 of 4 - See a demo at Strata Data Conference Booth 1201!
Matthew and Jack interned with Novetta’s Machine Learning Center of Excellence during the summer of 2019. This blog series discusses ADSynth, an app that creates a digital architecture diagram from a photo of a whiteboard sketch. As introduced in our first blog post, we built a simple iOS application alongside an AWS pipeline to provide an end-to-end solution to diagram creation. AWS was the obvious solution to build a truly scalable system capable of supporting multiple users creating diagrams at once. We were already using GPU enabled EC2 instances during the development of our object detection model, hence most…
ADSynth Blog Series: Defining the AWS Architecture Diagram
Part 3 of 4
Matthew and Jack interned with Novetta’s Machine Learning Center of Excellence during the summer of 2019. This blog series discusses ADSynth, an app that creates a digital architecture diagram from a photo of a whiteboard sketch. ADSynth leverages numerous image processing techniques to properly translate a whiteboard sketch into an editable, digital architecture diagram. These techniques are used to ensure that all detected components are connected and formatted correctly. This post describes how we utilized our YOLOv3 object detection model predictions in conjunction with the image of the whiteboard sketch to obtain a clean, presentable graphic ready for insertion…
ADSynth Blog Series: Object Detection Model and Training Data
Part 2 of 4
Matthew and Jack interned with Novetta’s Machine Learning Center of Excellence during the summer of 2019. This blog series discusses ADSynth, an app that creates a digital architecture diagram from a photo of a whiteboard sketch. We began with the backbone of ADSynth: the object detection model that makes diagram synthesis possible, as well as the data necessary to train the model. Implementation of YOLOv3 Architecture Based on the research we conducted on object detection, the architecture we decided to implement was YOLOv3. YOLOv3 is a deep neural network comprising of 106 layers and almost 63 million parameters. The…
ADSynth Blog Series: Digitizing Whiteboard Diagrams
Part 1 of 4
Matthew and Jack interned with Novetta’s Machine Learning Center of Excellence during the summer of 2019. This blog series discusses ADSynth, an app that creates a digital architecture diagram from a photo of a whiteboard sketch. This summer, we interned with Novetta’s Machine Learning Center of Excellence (ML COE). As computer science students with a passion for machine learning and problem solving, we were excited to have the opportunity to apply our classroom knowledge in an industry setting. The experience was rewarding, though transferring our theoretical understanding of machine learning to practice presented a steep learning curve (pun intended)….
Deploying Advanced Machine Learning Models at Scale
Novetta is using deep learning to automate advanced analytic pipelines for our customers. These new capabilities identify important pieces of information across large text, image, and video data sets, a challenge faced by many of our Defense and Intelligence Community customers. We developed Argo, a standardized, Docker-based model training and deployment pipeline, to enable rapid evaluation and deployment of state-of-the-art deep learning models. Argo enables models to be deployed as endpoints and, when used with Amazon SageMaker, it forms the foundation for us to deploy, scale, and manage our deep learning models. Simplifying infrastructure setup through Docker and AWS allows…
- 1
- 2
- 3
- 4
- Next Page »
