Automated feature learning: Deep learning algorithms automatically extract features from data, eliminating the need for hand-crafting.
Enhanced data handling: Can handle large and complex datasets effectively, leading to more robust models.
A stop sign can be shown as speed limit 40kmph.
Poisoning Attack: Attack during training phase
Attackers perturb the training set or the model
Insert malicious inputs in the training set (eg, that contain a trigger pattern to backdoor the model)
Change the labels to training inputs
Change the weights of a deployed trained model
Distribute poisoned pre-trained models to the public
The goal is to corrupt the ML model so that it performs incorrectly for some or all inputs
The adversary should obtain access to the training database or to the trained model
White box and Black box attacks
Adversarial attacks can be further classified into:
White box attack:
Aftackers have full knowledge about the ML model
I.e, they have access to parameters, hyperparameters, gradients, architecture, ete
Black box attack
Attackers don’t have access to the MI. model parameters, gradients, architecture
Attackers may query the black-box model (also known as the oracle) to obtain knowledge about the model
Black box attacks are more realistic, because model designers usually do not open source the model parameters
Non-target and Targets Attack
Non-targeted attack:
The goal is to mislead the classifier for an adversarial input to output any label other than the ground-truth label
E.g., perturb an image of a military tank, so that the model predicts it is any other class than a military tank
Targeted attack:
The goal is to mislead the classifier to predict a target label for an adversarial input
More difficult, in comparison to non-targeted attack
E.g. perturb an image of a turtle, so that the model predicts it is a riffle
How to find Adversial examples: intuition
Take an image x, which is labeled by the classifier (a deep model) as class y. le., C(x) – y
Create an adversarial imagé Xac by adding small perturbations & to the original image x, le., Kado – x + d, such that the distance D(x, Xad) – D(x, x 4d) is minimal
The aim for a non-targeted attack is that the classifier assigns a label to the adversarial image that is different than y, l.e, C(xav) – C(x+ б) / y
for a targeted attack, the aim is that
C(xadv) – C(x + d) = t
y, where t is the target class
Federated learning
Multiple hospitals are collaborated to work for cancer but due to HIPAA they can’t shared data but access. They can share by using Federated Learning. The weights of the machine model can be shared. Thism paradigm is called as Federated Learning.
Attack in Deep Federated Model
Various kinds of attacks in a federated learning paradigm, such as inference attacks, reconstruction attacks, poisoning attacks,
In inference attacks, the attacker can extract sensitive information about the training data from the learned features or parameters of the model, thus causing privacy issues.
Reconstruction attacks, on the other hand, try to generate the training samples using the leaked model parameters.
Poising attack in deep federated model
Poisoning attacke in a federated learning paradigo can ba categoried as data posoning ettacks or model posenira allacka
Both these attacks are designed to alter the behavior of the malicious client’s model
In data poisoning attacks, the attacker tries manipulating the training date by changing the ground truth label or carefully poisoning the existing data
la model poisoning attacks, the attacker aime to alter the model parameters or gradients before sending them to the global server.
Some clients try to decrease the accuracy of model to cause malicious attempts.
Cybersecurity: Protecting systems, networks and programs from digital attacks
If your organization experiences a cyberattack, sensitive information can be accessed, changed or destroyed, money can be extorted and/or business processes can be interrupted.
While sophisticated hacking is a valid threat to organizations, it is rarely the root cause of a data breach. The vast majority of data breaches and cybersecurity incidents are actually caused by a breakdown of basic cybersecurity processes and controls.
Cyber threats to watch out for in 2024
- Generative AI
- Third party
- Ransomware
Case Study – Al Deepfake Scam
Employee working out of the Company’s Hong Kong office received a message from the UK-based CFO, asking for a wire transfer to be made.
Employee was very suspicious and initially thought it was a phishing attack.
This employee was convinced to join a video call with the CFO and several other employees.
He recognized people on the call, so he made a payment of 25m dollars USD.
It turns out that everyone on the video call was Al generated, and the employee was talking with scammers the entire time.
Regulatory scrutiny will increase
Vendors are third party. There will be an increased focus on vendor risk management, as it pertains to cybersecurity. Do they have SOC2 report, do they have ISO27001 certificate?
There will be an increase in breaches and in the creativity of cyber attacks
Concern around the effectiveness of cyber prop forms to accurately present risk will continue
Digital Forensics Role in Cyber Security
Leveraging Digital Forensics in Mitigating Cyber Threats
Key Role of Digital Forensics in Countering Cyber Attacks
Malware Analysis
In-depth examination of malware to understand attack vectors and prevent future incidents.
Preserve Evidence
Digital forensics ensures the preservation of evidence critical for investigations and legal proceedings.
Identify Attack Origins
Through digital forensics, the origins of cyber attacks can be traced, enabling proactive security measures.
Collaborative Strategies for Cyber Security
The Vital Role of Collaboration in Countering Cyber Attacks
Emphasizing the importance of collaboration between digital forensics experts and law enforcement agencies
Cybersecurity Insights
Impact Assesment of Cyber Attacks
It is a process to determine the attack surface that an organization has and what impact will be caused if cyber attack occours.
Impact on Finance Sector
Uncover financial frauds causes and secure sensitive data
Security for Personal Data
Protect PIl, PIFI, PIHI and ensure data integrity
Government Sector Defense
Investigate breaches to safeguard national security
Forensic requirements
A forensic readiness plan is designed to prepare an organization for an unforeseen incidents and data losses.
An organization should review and evaluate security-technical controls, policies,
procedures, and skill sets
—as part of its preparation.
Staff members should be trained in incident response procedures so that they know their role in digital evidence processing and how important and sensitive it is in the event of an incident.
Due to constant monitoring and review, forensic readiness also helps ensure that employees comply with the organization’s policies and regulatory requirements.
Risk Mitigation
Enhances proactive identification and mitigation of cyber threats.
Legal Compliance
Ensures adherence to regulatory requirements and data protection laws.
Reputation Protection
Safeguards brand image and customer trust in the event of security breaches.
Incident Response
Improvement
Facilitates swift and effective responses to cyber incidents for damage control.
Cyber Investigations Challenges
Overcoming Obstacles in Cyber Investigations
Exploring Challenges Encountered in Digital Forensic Investigations
Volatile Digital Evidence
Ephemeral digital evidence poses challenges in preservation and analysis.
Data Access Restrictions
Restricted access to critical data impedes investigation progress.
Encryption Technologies
Encryption tools can hinder data decryption and analysis.
Enhanced Cybersecurity
Digital forensics significantly improves cybersecurity by providing insights into cyber incidents.
Challenges in Implementation
Implementing digital forensic strategies may face challenges such as complex data analysis and resource limitations
Best Practices Importance
Following best practices in digital forensics is crucial for effective cyber attack response and prevention.
Future Trends and Innovations
The future of digital forensics lies in advanced technologies like Al and machine learning for proactive threat dete
Encouraging Proactive Adoption organizations are encouraged to adopt digital forensic strategies proactively to bolster their cybersecurity posture.
Cyber Forensics
“The unique process of identifying, extracting, preserving, analyzing and presenting digital evidence in a legally acceptable manner.”
Data Centric 