Intelligent Autonomous Systems (IAS) reconstruct their perception through adaptive learning and meet mission objectives. IAS are highly cognitive, rich in knowledge discovery, reflective through rapid adaptation, and provide security assurance. It is paramount to have effective reasoning, decision-making, and understanding of operational context since IAS are exposed to advanced multi-stage attacks during training and inference time. Advanced malware types such as file-less malware with benign initial execution phase can mislead IAS to accept them as normal processes and execute malicious code later. IAS are also exposed to adaptive poisoning attacks where adversary inputs malicious data into training/testing set to manipulate the learning. Hence it is vital to monitor IAS activities/interactions to conduct forensics. This project will advance science of security in IAS through multifaceted advanced analytics, cognitive and adversarial machine learning, and cyber attribution based on the following approaches. (a) Implement deep learning-based application profiling to categorize adaptive cyberattacks and poison attacks on machine learning models using contextual information about the origin, trust, and transformation of data. (b) Using HW/OS/SW data to develop perception algorithms using LSTM deep neural networks for detecting malware/anomalies and classifying dynamic attack contexts. (c) Facilitate cyber attribution for forensics through privacy-preserving provenance structure for knowledge representation and perform intrusion detection sampling on HW /OS/SW data. (d) Employ advanced data analytics to aid ontological and semantic reasoning models to enhance decision-making, attack adaptiveness, and self-healing.
International Semantic Intelligence Conference (ISIC 2021), Feb 25-27, 2021, New Delhi, India EMAIL: bbshail@purdue.edu (A. 2); ️© 2020 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
Our focus is on constraints, barriers and challenges such as poorly understood attack surfaces, data set training availability and biases, processing latency, human understanding of AI results, AI/ML countermeasures, human-machine disparity, measurement of effects. We propose novel approaches for privacy-preserving cyber attribution, intrusion detection, adversarial machine learning, malware/anomaly detection, reasoning, and decision-making. Cyber attribution involves extracting software, hardware, and operating system data to perform intrusion detection sampling (fixed or dynamic sampling), generating efficient provenance structure that is populated with specific data required for a particular analysis or learning, and labeling and tagging to properly represent the information obtained. The processed data is distributed to the cognitive module where the data is checked for any malicious data presence through poison attack filter. The filtered data is transmitted to cognitive computing module and knowledge discovery module, where the data is fed into supervised, unsupervised, and LSTM models to perform learning and advanced analytics. Based on multifaceted dimensions of data analytics, reasoning and decision-making ability of IAS are enhanced. The overall architecture of the proposed model-secure intelligent autonomous systems with cyber attribution-is demonstrated in figure 1. Intelligent autonomous systems receive large amounts of diverse data from various data sources. In addition, they operate in a dynamic operational context and interact with numerous entities such as other TAS, UAVs, satellites, sensors, cloud systems, analysts, malicious actors, and compromised systems.
Cyber attribution module constitutes a stream data processor where data streams are labeled / tagged on-the-fly for better knowledge representation and categorization. This data is stored as monitored or provenance data with its origin and historical information. For preserving privacy, detailed provenance data is reduced in its scope to include only necessary data for a particular analysis or learning. This module uses Provenance Ontology (PROV-O) structure (elaborated in a later section) to obscure unnecessary or privacy-compromising data. Furthermore, the attribution model monitors data generated by software (application parameters), hardware (memory bytes and instructions), and operating system (system calls). This data is used to conduct periodic sampling to identify signatures of intrusion activities. Once the data is processed, it goes through adversarial machine learning model. Attackers can insert malicious data into training and testing dataset to influence machine learning models. In order to isolate poisonous data, poison data filter performs methods such as classification of verified and unverified data as well as outlier extraction. Once the poisonous data is removed the data (raw or provenance data) is sent to Cognitive computing module and Knowledge discovery module. In Cognitive computing module, depends on the data and efficiency of machine learning methods, malware / anomaly detection is performed through either deep learning methodologies such as Long short-term memory (LSTM) e.g. Recurrent Neural Networks (RNN) or Convolutional Neural Networks (CNN) or light-weight yet powerful machine learning methods such as Support Vector Machines (SVM), Random Forests (RF), and K-Nearest Neighbors (KNN). In addition, cognitive computing module consists of reasoning engine, which is driven by rule sets, semantic, and ontological reasoning. Both anomaly detection module and reasoning engine module influence the attack adaptiveness (reflexivity) and self-healing of IAS, where decisions obtained through reasoning and learning are turned into actions. With this extensive cognitive computing modules, the final response from IAS to other interacting entities will be a secure and trusted one.
Knowledge discovery module facilitates multi-faceted dimensions of advanced data analytics including regression analysis, supervised learning, unsupervised learning, and pattern-recognition. Discovered knowledge is shared with cognitive computing module for further learning. The proposed structure provides robust cyber resilience and autonomous operation of the system.
Cognitive computing is a vital part of
security in autonomous systems. In particular,
malware and anomaly detection has become a
biggest challenge with increase in
sophistication in attacks such as file-less
malware [
Decentralized machine learning is a promising emerging paradigm in view of global challenges of data ownership and privacy. We consider learning of linear classification and regression models, in the setting where the training data is decentralized over many user devices, and the learning algorithm must run on device, on an arbitrary communication network, without a central coordinator. We plan to utilize and advance COLA, a new decentralized training algorithm [23] with strong theoretical guarantees and superior practical performance. This framework overcomes many limitations of existing methods, and achieves communication efficiency, scalability, elasticity as well as resilience to changes in data and participating devices. We will consider fault tolerance to dropped and oscillation of nodes from connected to disconnected and attacks on the nodes. The learning has to be communicationefficient decentralized framework and free of parameter tuning. COLA offers full adaptively to heterogeneous distributed systems on arbitrary network topologies and is adaptive to changes in network size and data and offers fault tolerance and elasticity. IAS should have clear understanding of its operational context, it's won processes, and its interactions with neighboring entities. In this project, the cognitive computing module consists of three major components: (1) Malware / anomaly detection module, (2) Reasoning engine, and ( 4) Reflexivity engine. Cyber attribution data (system monitoring data or provenance data) is sent to cognitive computing engine for analysis where the system profiles the applications based on machine learning models. In this paper, we will focus on the cognitive autonomy property of the autonomous systems.
We use instruction sequences executed in memory by application to understand the behavior of each application.
Input: n-gram sequences of instructions from memory Output: Binary classification of benign or malicious Step 1: Define a finite set I of instructions {i1, i2, ..., in} in the system. Instructions are executed based on time epochs i.e., timeseries data. Step 2: Given an observed sequence of {i1, i2, ..., in}, we find the set N of the top P sequences to be executed at time t. The size of the set N varies in each prediction and is determined by ngrams input as well as the clusters in the output of the model. Step 3: At time t, the sequence {i1, i2, ..., in} is benign if i1 is in P, otherwise malicious.
Algorithm 1: Application Behavioral
Profiling Algorithm
Advanced malware such as ransomware
encrypts IAS data without authorization. Since
it does not alter the system configurations and
leave a footprint, it is difficult to detect them.
But based on the executed instruction
sequences and constants (also known as magic
constants) used for encryption mechanism
during malware execution, applications can be
profiled. First, we will sample the address
spots for every 1,000,000 instructions (fixed
sampling). After a fixed period of time, we
will calculate the frequently occurring
addresses and their relevant process ids. A
threshold T will be set for data extraction. For
example, extract memory bytes and
instructions from top T = 10% of the global
list of sampled addresses (sorted in descending
order based on their frequency of occurrence).
Once opcode and memory bytes data is
collected, we will extract features such as
ngram, bigram, unigram features, magic
constants feature, cosine similarity with
instructions occurrences, and standard
deviation. Cosine similarity metric is one of
the most efficient method to learn from large
datasets [20]. It plays a crucial role in
understanding similarity between two feature
vectors when the magnitude of the vector is
large or unspecified
i.e., it can either be unigram, bigram, or
ngram features. Given two feature vectors Vi =
{f11, f12, ...} and Vi = {f21, f22, ...}, where f11,
f21, . . .are values of a particular feature, the
cosine similarity is given as,
The cosine similarity lies between O and 1. If
the orientation of the two feature vectors is the
same then the similarity between them is Cos
O = 1 i.e., there is zero angle between them.
But when the angle is 90° (the orientation of
the feature vectors is at an angle of 90) then
the
similarity is Cos 90 = 0. The similarity score
varies between [O, ½). Once the features are
extracted, we will implement RF, SVM, and
KNN learning models. K-NN is one of the
simplest yet powerful classifier with high
computational efficiency as well as accuracy
[
We presented two approaches for detecting through profiling evasive malware applications. We use both light-weight machine learning models as well as deep learning models to profile and understand the behavior of autonomous systems. This multimodel approach is advantages when it comes to computational resources in mission critical systems. Based on the data and sample size, appropriate model can be selected for analysis. In particular, light-weight machine learning models use less computational resources and they have considerably less time complexity. On the other hand, LSTM model can provide robust classification with fundamental data, which enables IAS to understand evasive malware at basic level.
This research is funded by Northrop Grumman Corporation.