Systems and methods for applying reinforcement learning to cybersecurity graphs
Methods, systems and computer program products are provided for integrating risk and threat intelligence from various sources, to provide real-time awareness of potential threats to a computer network, which are now described herein in terms of an example enterprise system.
1. A cybersecurity method for determining exploits, comprising the steps of:
receiving a graph representing a digital network of a plurality of nodes forming a federated learning network, the graph including at least one vulnerability for each of the plurality of nodes;
receiving, for the plurality of nodes, a plurality of embeddings based on the graph, wherein the plurality of embeddings include a vector of real numbers representing the plurality of notes in the graph;
assigning an agent an initial node from the plurality of nodes;
querying the graph to obtain a plurality of accessible nodes and at least one vulnerability for the accessible nodes;
determining a transition for the agent to take from the initial node to a next accessible node from the plurality of accessible nodes;
computing using a neural network, a reward for moving to the next accessible node;
assigning the agent a new state corresponding to the next accessible node;
collecting, by a collected experience database, a history of node assignments of the agent, a plurality of connections taken by the agent, and a plurality of rewards the agent received for transitioning across the plurality of connections;
updating a plurality of parameters of a neural network using the data collected by the collected experience database, wherein the information collected by each of a plurality of agents is used to further update the plurality of parameters of the neural network while not sharing graph data contributed by a plurality of graphs; and
determining, by the agent, what action from a plurality of available actions to take next using the neural network.
2. The method according to claim 1 , wherein determining the transition for the agent to take is based on the embeddings of a plurality of connections between the initial node and the plurality of accessible nodes.
3. The method according to claim 1 , further comprising the step of:
determining from the plurality of nodes on the graph at least one node having a vulnerability score relatively higher than a plurality of vulnerability scores of other nodes in the graph.
4. The method according to claim 1 , further comprising the step of:
detecting an exploit associated with a node in the graph representing the digital network; and
preventing the agent from traversing the node in the digital network that is associated with the exploit and represented in the graph, thereby inhibiting the exploit.
5. The method according to claim 1 , further comprising the step of:
determining a starting point in the graph that can be exploited to gain access to another node in the graph.
6. A non-transitory computer-readable medium having stored thereon sequences of instructions, the sequences of instructions including instructions which when executed by a computer system causes the computer system to perform determining exploits by:
receiving a graph representing a digital network of a plurality of nodes forming a federated learning network, the graph including at least one vulnerability for each of the plurality of nodes;
receiving for the plurality of nodes, a plurality of embeddings based on the graph;
assigning an agent an initial node from the plurality of nodes;
querying the graph to obtain a plurality of accessible nodes and at least one vulnerability for the accessible nodes, wherein the plurality of embeddings include a vector of real numbers representing the plurality of nodes in the graph;
determining a transition for the agent to take from the initial node to a next accessible node from the plurality of accessible nodes;
computing a reward for moving to the next accessible node;
assigning the agent a new state corresponding to the next accessible node;
collecting a history of node assignments of the agent, a plurality of connections taken by the agent, and a plurality of rewards the agent received for transitioning across the plurality of connections;
updating a plurality of parameters of a neural network using the data collected by the collected experience database, wherein the information collected by each of a plurality of agents is used to further update the plurality of parameters of the neural network while not sharing graph data contributed by a plurality of graphs; and
determining, by the agent, what action from a plurality of available actions to take next using the neural network.
7. The non-transitory computer-readable medium according to claim 6 , wherein determining the transition for the agent to take is based on the embeddings of a plurality of connections between the initial node and the plurality of accessible nodes.
8. The non-transitory computer-readable medium of claim 6 , further having stored thereon a sequence of instructions for causing the one or more processors to perform:
determining from the plurality of nodes on the graph at least one node having a vulnerability score relatively higher than a plurality of vulnerability scores of other nodes in the graph.
9. The non-transitory computer-readable medium of claim 6 , further having stored thereon a sequence of instructions for causing the one or more processors to perform:
detecting an exploit associated with a node in the graph representing the digital network; and
preventing the agent from traversing the node in the digital network that is associated with the exploit and represented in the graph, thereby inhibiting the exploit.
10. The non-transitory computer-readable medium of claim 6 , further having stored thereon a sequence of instructions for causing the one or more processors to perform:
determining a starting point in the graph that can be exploited to gain access to another node in the graph.
11. A cybersecurity system for determining exploits, comprising:
one or more processors configured to:
generate, for a plurality of nodes, a plurality of embeddings based on a graph representing a digital network of the plurality of nodes forming a federated learning network, the graph including at least one vulnerability for each of the plurality of nodes, wherein the plurality of embeddings include a vector of real numbers representing the plurality of nodes in the graph;
assign an agent an initial node from the plurality of nodes;
query the graph to obtain a plurality of accessible nodes and at least one vulnerability for two or more accessible nodes;
determine a transition for the agent to take from the initial node to a next accessible node from the plurality of accessible nodes;
compute, using a neural network, a reward for moving to the next accessible node;
assign the agent a new state corresponding to the next accessible node;
a collected experience database configured to collect a history of node assignments of the agent, a plurality of connections taken by the agent, and a plurality of rewards the agent received for transitioning across the plurality of connections;
the one or more processors further configured to update a plurality of parameters of a neural network, wherein the information collected by each of a plurality of agents is used to that further update the plurality of parameters of the neural network while not sharing graph data contributed by a plurality of graphs; and
the agent further configured to determine what action from a plurality of available actions to take next using the neural network.
12. The cybersecurity system according to claim 11 , wherein determining the transition for the agent to take is based on the embeddings of a plurality of connections between the initial node and the plurality of accessible nodes.
13. The cybersecurity system according to claim 11 , wherein the one or more processors are further configured to determine, from the plurality of nodes on the graph, at least one node having a vulnerability score relatively higher than a plurality of vulnerability scores of other nodes in the graph.
14. The cybersecurity system according to claim 11 , wherein the one or more processors are further configured to:
detect an exploit associated with a node in the graph representing the digital network; and
prevent the agent from traversing the node in the digital network that is associated with the exploit and represented in the graph, thereby inhibiting the exploit.
15. The cybersecurity system according to claim 11 , wherein the one or more processors are further configured to:
determine a starting point in the graph that can be exploited to gain access to another node in the graph.