<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Umanath Nayak on Medium]]></title>
        <description><![CDATA[Stories by Umanath Nayak on Medium]]></description>
        <link>https://medium.com/@umanath.nayak1?source=rss-4406b526148e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ADRdfOB8ZLtsRqiFokWv_A.jpeg</url>
            <title>Stories by Umanath Nayak on Medium</title>
            <link>https://medium.com/@umanath.nayak1?source=rss-4406b526148e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 03:20:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@umanath.nayak1/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Postgres & AI Extensions]]></title>
            <link>https://medium.com/@umanath.nayak1/postgres-ai-extensions-37e75820b7c8?source=rss-4406b526148e------2</link>
            <guid isPermaLink="false">https://medium.com/p/37e75820b7c8</guid>
            <dc:creator><![CDATA[Umanath Nayak]]></dc:creator>
            <pubDate>Sun, 20 Jul 2025 18:33:40 GMT</pubDate>
            <atom:updated>2025-07-20T23:29:07.819Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CfTeEalOjMCVLm7tLZupVQ.jpeg" /><figcaption>image credit:: meta.ai</figcaption></figure><p>In the last month, both Databricks and Snowflake announced large acquisitions of Postgres product companies . Databricks acquired Neon, a serverless PostgreSQL provider, while Snowflake acquired Crunchy Data, an enterprise-grade PostgreSQL provider. Both these acquisitions highlight the growing importance of Postgres and the enablement of this technology in the Data, AI and other agentic AI use cases.</p><p>In a 2023 popularity survey of 90000+ developers on stack overflow, Postgres was ranked on the top.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5DlN9HE8kHyt6RAacvqEkQ.png" /></figure><p><strong>Background of Postgres</strong></p><p>Postgres was developed at the University of California Berkeley in the 1980s as a successor to Ingres, another Database product at that time . In the 1990s Postgres changed its name to PostgreSQL , highlighting the relational and to reflect its support for SQL ( Structured Query Language)</p><p><strong>Postgres Architecture</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/581/1*g6Oe91i3Qs3bRwYlb11iVA.png" /></figure><p>Like most traditional RDBMSs, Postgres follows a Process + Memory + Storage architecture. The Memory areas provide caching for frequently used blocks and for write ahead log buffers ( used for point-in-time recovery or instance recovery ) and the storage layer comprises of data files and log files.</p><ol><li><strong>Process architecture</strong></li></ol><ul><li>Postgres Server Process (formerly Postmaster): This is the parent process that starts first, listens for incoming connections, and initializes shared memory. It’s responsible for managing client connections and spawning other processes.</li><li>Backend Processes: For each client connection, the Postgres server process forks a new, independent backend process to handle that session. This backend process is responsible for all communication and query execution for that specific client.</li></ul><p>Background Processes: PostgreSQL runs several background processes to manage tasks not directly related to client sessions. These include:</p><ul><li>Checkpointer: Periodically flushes dirty pages from shared buffers to disk, ensuring data durability and bounding recovery time after a crash.</li><li>WAL Writer: Writes Write-Ahead Log (WAL) changes from memory (WAL buffer) to disk, guaranteeing that changes are logged before being applied to the data files, crucial for crash recovery and replication.</li><li>Autovacuum Workers: Automatically cleans up “dead” tuples (old versions of rows) and updates statistics, preventing performance degradation and reclaiming disk space.</li></ul><p>2. <strong>Memory management</strong></p><p>Shared Memory — A crucial memory component shared by all PostgreSQL processes. It contains:</p><ul><li>Shared Buffers: Caches data blocks read from disk, significantly reducing disk I/O and improving query response times.</li><li>WAL Buffer: Temporarily stores database changes (WAL records) before they are written to disk, ensuring data durability.</li><li>Local Memory: Each backend process utilizes local memory structures, such as work_mem, for specific operations like sorting and hashing during query execution.</li></ul><p>3. <strong>Data storage and retrieval</strong></p><ul><li>Database files : A collection of databases managed by a single PostgreSQL server.</li><li>Databases: Within a cluster, databases are the primary containers for data. Each database has its own directory on disk.</li></ul><p><strong>Postgres vs Other Relational Products</strong></p><p>This table gives a comparison of how PostgreSQL compares with other competitor relational products in terms of popularly used features</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z9qIPgtwmUGfBKTfiuYPIw.png" /><figcaption>Feature comparison of Postgres vs other Relational Products</figcaption></figure><p><strong>Postgres AI features</strong></p><p>Both Snowflake and Databricks have carried out Postgres based acquisitons to enable strategic AI features on their Data Clouds.</p><p><strong>pgvector Extension</strong>: This open-source extension is crucial for AI in PostgreSQL. It allows storing, querying, and indexing high-dimensional vector data directly within the database. This is particularly useful for applications involving:</p><ul><li>Similarity search: Finding data points (e.g., images, text, user profiles) that are similar to a given input based on their vector representations.</li><li>Recommendation systems: Suggesting items or content based on user preferences or item similarities.</li><li>Natural language processing (NLP): Enhancing text search relevance and enabling contextual understanding of queries.</li><li>Retrieval Augmented Generation (RAG): Improving the retrieval of relevant information for generating responses by AI models.</li></ul><p>pgvector extensions give Postgres users an ability to use semantic and vector search for RAG enabled queries enabled by LLMs. for e.g if someone using a certain genre of movies , vector search will help make the search closer to the preferred genre of the viewer and can make recommendations on what to watch next.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/594/1*kh0IdAXP91zTWuF7M_x9Yw.png" /></figure><p><strong>Examples of using pgvectors</strong></p><ol><li>Creating a table with a vector column</li></ol><pre>SQL<br>-- Create a table with a vector column<br>CREATE TABLE items (<br>    id SERIAL PRIMARY KEY,<br>    name VARCHAR(255),<br>    embedding VECTOR(3)  -- 3-dimensional vector<br>);<br><br>-- Insert some sample data<br>INSERT INTO items (name, embedding) VALUES<br>    (&#39;Item 1&#39;, &#39;[1, 2, 3]&#39;),<br>    (&#39;Item 2&#39;, &#39;[4, 5, 6]&#39;),<br>    (&#39;Item 3&#39;, &#39;[7, 8, 9]&#39;);</pre><p>2. Querying similar vectors</p><pre>SQL<br>-- Find the nearest neighbors to a given vector<br>SELECT name, embedding &lt;-&gt; &#39;[1, 2, 3]&#39; AS distance<br>FROM items<br>ORDER BY distance<br>LIMIT 5;</pre><p>In this example, the &lt;-&gt; operator calculates the Euclidean distance between two vectors.</p><p>3. Using a Python client to work with pgvector</p><pre>import psycopg2<br>from psycopg2.extras import DictCursor<br><br># Connect to the database<br>conn = psycopg2.connect(<br>    host=&quot;localhost&quot;,<br>    database=&quot;mydatabase&quot;,<br>    user=&quot;myuser&quot;,<br>    password=&quot;mypassword&quot;<br>)<br><br># Create a cursor<br>cur = conn.cursor(cursor_factory=DictCursor)<br><br># Insert a new item with a vector embedding<br>new_item_name = &quot;Item 4&quot;<br>new_item_embedding = [10, 11, 12]<br>cur.execute(&quot;INSERT INTO items (name, embedding) VALUES (%s, %s)&quot;, (new_item_name, new_item_embedding))<br><br># Query similar vectors<br>query_vector = [1, 2, 3]<br>cur.execute(&quot;SELECT name, embedding &lt;-&gt; %s AS distance FROM items ORDER BY distance LIMIT 5&quot;, (query_vector,))<br>results = cur.fetchall()<br><br># Print the results<br>for row in results:<br>    print(f&quot;Name: {row[&#39;name&#39;]}, Distance: {row[&#39;distance&#39;]}&quot;)<br><br># Close the cursor and connection<br>cur.close()<br>conn.close()</pre><p>4: Generating embeddings with a library like Transformers</p><pre>from transformers import AutoModel, AutoTokenizer<br>import torch<br><br># Load a pre-trained model and tokenizer<br>model_name = &quot;sentence-transformers/all-MiniLM-L6-v2&quot;<br>model = AutoModel.from_pretrained(model_name)<br>tokenizer = AutoTokenizer.from_pretrained(model_name)<br><br># Define a function to generate embeddings<br>def generate_embedding(text):<br>    inputs = tokenizer(text, return_tensors=&quot;pt&quot;)<br>    outputs = model(**inputs)<br>    embeddings = outputs.last_hidden_state[:, 0, :]<br>    return embeddings.detach().numpy()[0]<br><br># Generate an embedding for a piece of text<br>text = &quot;This is a sample piece of text.&quot;<br>embedding = generate_embedding(text)<br>print(embedding)</pre><p><strong>pgai extension</strong></p><p>Pgai provides functions to easily integrate AI models with your PostgreSQL database, particularly for tasks like embedding generation, summarization, and RAG (Retrieval Augmented Generation). Here are some code examples</p><p>These examples demonstrate the AI advancements in open source Postgres both on a query use case or an extension of RAG for vector databases.</p><pre>-- install pgai (also installs pgvector via cascade)<br>CREATE EXTENSION IF NOT EXISTS ai CASCADE;<br><br>-- a table for storing movie viewing log history<br>CREATE TABLE movie_viewing_log (<br>    id int NOT NULL PRIMARY KEY,<br>    movie text,<br>    viewer_id,<br>    genre,<br>    &quot;date&quot; timestamptz,<br>    summary text,<br>    detail text<br>);<br><br>-- (load your event data, for example, using \copy command in psql)<br>\copy movie_viewing_log from &#39;movie_view_log.csv&#39; with (format csv)<br><br>-- generate embeddings for each row and store them in a new table<br>INSERT INTO movie_view_embed (id, embedding)<br>SELECT id, ai.openai_embed(&#39;text-embedding-3-small&#39;, &quot;movie&quot;)<br>FROM movie_viewing_log;</pre><pre>-- we want to embed each row of movie viewing history and put the embedding in this table<br>create table movie_view_embed<br>( id int not null primary key<br>, embedding vector(1536) -- the vector type comes from the pgvector extension<br>);<br><br>-- select from the first table, embed the content, and insert in the second table<br>insert into movie_view_embed (id, embedding)<br>select<br>  id<br>, openai_embed<br>  ( &#39;text-embedding-3-small&#39;<br>    -- create a single text string representation of the movie viewing<br>  , format(&#39;date: %s movie: %genre %s summary: %s detail: %s&#39;, &quot;date&quot;, &quot;movie&quot;,&quot;genre&quot;, summary, detail)<br>  ) as embedding<br>from movie_viewing_log<br>;</pre><pre>-- summarize and categorize movies to make viewer recommendations<br>select openai_chat_complete<br>( &#39;gpt-4o&#39;<br>, jsonb_build_array<br>  ( jsonb_build_object<br>    ( &#39;role&#39;, &#39;system&#39;<br>    , &#39;content&#39;, &#39;You are a AI agent who summarizes movie viewership details per viewer &#39;<br>    )<br>  , jsonb_build_object<br>    ( &#39;role&#39;, &#39;user&#39;<br>    , &#39;content&#39;<br>    , -- build up a list of the movie viewing details to append to the prompt<br>      concat<br>      ( E&#39;Summarize the following list of movie views from October 2024 in a markdown format.\n\n&#39;<br>      , string_agg(x.movie_desc, E&#39;\n\n&#39;)<br>      )<br>    )<br>  )<br>)-&gt;&#39;choices&#39;-&gt;0-&gt;&#39;message&#39;-&gt;&gt;&#39;content&#39;<br>from<br>(<br>    -- convert each to a text format<br>    select format<br>    ( E&#39;%s %s\n\tmovie: %s\n\tgenre: %s\n\tdate: %s\n\tdetail: %s&#39;<br>    , row_number() over (order by &quot;date&quot;)<br>    , summary<br>    , &quot;movie&quot;<br>    , &quot;genre&quot;<br>    , &quot;date&quot;<br>    , detail<br>    ) as movie_desc<br>    from movie_viewing_log<br>    --look at movie views from October 2024<br>    where date_trunc(&#39;month&#39;, &quot;date&quot;) = &#39;2024-10-01 00:00:00+00&#39;::timestamptz<br>    order by &quot;date&quot;<br>) x<br>;</pre><p>With these extensions, an open source product like Postgres will build upon many use cases for AI based Data querying and searches.</p><p><strong>Public Cloud hosting options for Postgres</strong></p><p>Public cloud providers like AWS, Google Cloud, and Azure offer managed PostgreSQL services, simplifying database setup, operation, and scaling. These providers, often referred to as the “Big Three,” provide robust and scalable PostgreSQL solutions. Other options include dedicated PostgreSQL services like ElephantSQL, Aiven, and Render, as well as general-purpose platforms like Heroku and DigitalOcean that offer PostgreSQL add-ons.</p><p>Major Cloud Providers (The Big Three):</p><p>Amazon Web Services (AWS):</p><ul><li>Amazon RDS for PostgreSQL provides a fully managed service for setting up, operating, and scaling PostgreSQL deployment</li></ul><p>Microsoft Azure:</p><ul><li>Azure Database for PostgreSQL provides a managed service for running and scaling PostgreSQL databases.</li></ul><p>Google Cloud Platform(GCP):</p><ul><li>Google Cloud SQL for PostgreSQL offers a scalable and secure cloud-based solution.</li></ul><p><strong>Further Reading</strong></p><ul><li>PostgreSQL 12 Official Documentation, <a href="https://www.postgresql.org/docs/">https://www.postgresql.org/docs/</a></li><li>PostgreSQL for Data Architects, Jayadevan Maymala, Packt Publishing, March 2015</li><li>PostgreSQL 10 High Performance, Enrico Pirozzi, Gregory Smith, Ibrar Ahmed, April 2018</li><li>Learning PostgreSQL 11, Andrey Volkov, Salahaldin Juba, January 2019</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=37e75820b7c8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Role of AI in Cybersecurity]]></title>
            <link>https://medium.com/@umanath.nayak1/the-role-of-ai-in-cybersecurity-a6ec1c32125c?source=rss-4406b526148e------2</link>
            <guid isPermaLink="false">https://medium.com/p/a6ec1c32125c</guid>
            <dc:creator><![CDATA[Umanath Nayak]]></dc:creator>
            <pubDate>Sun, 29 Jun 2025 19:17:11 GMT</pubDate>
            <atom:updated>2025-06-29T19:17:11.163Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VfUxo34cxXb_kOUCNcsOtg.jpeg" /><figcaption>image credit::meta.ai</figcaption></figure><p><strong>What is Cybersecurity?</strong></p><p>Cybersecurity is a practice to protect systems, data , information and organizational sensitive information from malicious cyberattacks and cyberattackers. It also includes measures to prevent threats like hacking, malware, phishing and ransomware. There have been many instances where organizations facing these attacks have faced significant financial or worse, reputational harm.</p><p>There are regulatory standards and frameworks available on how to follow and implement multi-layered systems to prevent cyberattacks . Organizations can implement these standards to strengthen their cybersecurity posture:</p><p><a href="https://www.nist.gov/cyberframework">Cybersecurity Framework</a></p><p>Also , depending on the industry there are specific standards available like PCI-DSS ( Payments ) and HIPAA ( Healthcare ).</p><p><strong>Types of Cybersecurity Threats</strong></p><p>Cybersecurity threats can be broadly categorized into 2 areas:</p><ol><li><strong>Enterprise Level Threats</strong></li></ol><ul><li>Advanced Persistent Threats ( APTs ) : These are sophisticated, targeted attacks by organized groups to steal sensitive data or disrupt operations</li><li>Business email compromises (BEC ): These are targeted phishing attacks on executives or finance teams to trick them into transferring funds or revealing sensitive information</li><li>Supply Chain attacks : These are compromising third-party vendors or partners to gain access to an organization’s systems or data.</li><li>Ransomware Attacks on Critical Infrastructure: Targeting hospitals, power plants, or other critical infrastructure with ransomware, potentially disrupting operations and putting lives at risk.</li><li>Cloud Security Breaches: Unauthorized access to cloud-based data or applications, potentially exposing sensitive information.</li><li>Data Exfiltration: Unauthorized transfer of sensitive data outside the organization, often through encrypted channels.</li></ul><p>2. <strong>Personal Cybersecurity Threats</strong></p><ul><li>Identity Theft: Someone stealing your personal info to open credit cards, loans, or access your accounts</li><li>Social Media Hacking: Someone gaining unauthorized access to your social media accounts to steal info or spread malware.</li><li>Online Banking Threats: Phishing, malware, or other attacks targeting online banking accounts to steal financial info.</li><li>Password Theft: Someone stealing or cracking your passwords to access sensitive info or accounts.</li><li>Device Theft or Loss: Losing or having your device stolen, potentially exposing sensitive data.</li><li>Phishing and Smishing: Receiving fake emails, texts, or messages that trick you into revealing sensitive info.</li><li>Online Scams: Falling victim to fake online offers, contests, or schemes that steal money or info</li><li>Data Breaches: Your personal data being exposed in a breach of a company or organization’s security</li></ul><p><strong>Real-life examples of recent Cybersecurity attacks</strong></p><ul><li>Ransomware attacks: Companies like Coca-Cola and MathWorks have recently been targeted by ransomware attacks.</li><li>Data Leaks:The Open Credentials Database suffered a data leak due to an unsecured server, showcasing the risks of unsecured data storage</li><li>Third Party Breaches:Adidas was impacted by a breach at a third-party vendor, demonstrating the importance of securing the entire supply chain.</li><li>Snowflake Data Breach:Ticketmaster and other organizations experienced data theft due to a breach at Snowflake, a cloud storage company.</li><li>Phishing Attacks:Phishing attacks, where malicious actors use deceptive emails or websites to steal information, remain a prevalent and disruptive threat.</li></ul><p><strong>AI in Cybersecurity</strong></p><p>Since the launch of OpenAI’s ChatGPT in 2022, there has been an exponential increase in the use cases across many industries on how to leverage from LLM( Large Language Model) based AI for simple to multi-modal complex use cases. Before the launch of ChatGPT, Machine Learning had limited use cases from translation to serialized and constrained problem solving . The transformer based models running on high performance GPUs brought a paradigm shift by predicting “what-if” and “what-next” scenarios via newly launched open source and closed source models. Advanced LLMs are today able to navigate complex tasks and control physical devices such as robots and cars with very high precision. Models are also evolving daily by training with new data and faster compute and can adapt to real-life use cases by “human-in-the-loop” interactions and reinforced learning.</p><p>With newer AI models, AI Agents and AgenticAI evolving , organizations can defend against cyber threats . Models can leverage machine learning, deep learning and data analytics to protect digital systems and networks.</p><p>There are many use cases that AI can help address:</p><ol><li>Threat Detection and Response : AI models have the ability to get trained on vast amount of logs and training data to identify malicious patterns and anomalies , predicting threats before they occur</li><li>AI based Incident response : AI can automate the repetitive tasks in incident response, such as threat analysis, containment and remediation, freeing up Incident monitoring teams to focus on more complex challenges</li><li>Observability and Self-Healing:AI systems can continuously monitor networks and systems, providing round-the-clock protection and detecting threats even outside of business hours.</li><li>Vulnerability Management:AI can be used to identify and prioritize vulnerabilities in systems and applications, helping organizations proactively address potential weaknesses.</li></ol><p><strong>Machine Learning Algorithms for Cybersecurity</strong></p><p>Machine Learning can help significantly with early threat detection and response. Some of the algorithms that can help with this are:</p><p><strong>Anomaly Detection</strong></p><ul><li>Local Outlier Factor (LOF): Identifies anomalies in network traffic or system behavior.</li><li>One-Class SVM (Support Vector Machine): Detects unusual patterns in data.</li><li>Isolation Forest: Isolates anomalies in high-dimensional data.</li></ul><p><strong>Classification</strong></p><ul><li>Random Forest: Classifies network traffic or system behavior as malicious or benign.</li><li>Gradient Boosting: Classifies threats, such as malware or phishing attacks.</li><li>Neural Networks: Classifies complex patterns in data.</li></ul><p><strong>Clustering</strong></p><ul><li>K-Means: Groups similar network traffic or system behavior for anomaly detection.</li><li>DBSCAN (Density-Based Spatial Clustering of Applications with Noise): Clusters data points for anomaly detection.</li></ul><p><strong>Predictive Analytics</strong></p><ul><li>Recurrent Neural Networks (RNNs): Predicts future threats or anomalies based on historical data.</li><li>Long Short-Term Memory (LSTM) Networks: Predicts sequences of events, such as network traffic patterns.</li></ul><p><strong>Real Life examples of AI in Cybersecurity</strong></p><p>1. Malware Detection and Classification:</p><ul><li>Traditional anti-virus vs. ML-powered solutions: Traditional antivirus software relies on known malware signatures, which is effective against known threats but struggles with new or evolving malware.</li><li>Machine Learning approach: ML-powered solutions analyze file behavior, characteristics, and execution patterns to detect malicious software, even for zero-day threats (previously unknown threats).</li><li>Example: Cylance, a software company, developed a smart antivirus that uses machine learning to detect viruses or malware based on their behavior, not just their signature. CrowdStrike’s Falcon platform also uses AI to detect malware.</li></ul><p>2. Intrusion Detection Systems (IDS):</p><ul><li>ML enhancing IDS: Machine learning significantly improves the capabilities of intrusion detection systems (IDS).</li><li>Anomaly detection: ML algorithms analyze network traffic and user behavior to establish baseline patterns and identify deviations, which can indicate potential security breaches.</li><li>Example: Darktrace uses unsupervised ML to detect anomalies in network traffic and alert businesses to potential breaches. Exabeam’s platform utilizes AI-driven detections to identify high-risk threats by learning user and entity behavior.</li></ul><p>3. Spam and Phishing Detection:</p><ul><li>ML for email analysis: Machine learning algorithms and Natural Language Processing (NLP) techniques analyze email content, sender behavior, and URL patterns to identify and filter phishing attempts.</li><li>Example: An email monitoring system project can be created to scan employee emails and look for characteristics that might indicate a phishing attempt.</li></ul><p>4. User and Entity Behavior Analytics (UEBA):</p><ul><li>Detecting insider threats: ML helps UEBA systems establish baselines of normal user and system behavior, then flag unusual activities that could indicate a threat, such as unauthorized access or unusual downloads.</li><li>Example: Machine learning can detect unusual login times or inconsistent remote access to detect potential account compromises.</li></ul><p>5. Credit Card Fraud Detection:</p><ul><li>Analyzing transaction data: ML algorithms analyze transaction data in real time to detect unusual spending patterns that might indicate fraud.</li><li>Example: A credit card fraud detection project uses machine learning techniques like K-nearest neighbors, Random Forest, and decision trees to build classification models for analyzing transactions</li></ul><p><strong>Python ML libraries for enabling AI in Cybersecurity</strong></p><p>Python, with its extensive libraries and frameworks has become a popular choice in enabling AI based solutions in Cybersecurity</p><p><strong>Machine Learning Libraries</strong></p><p>These popular Python ML libraries provide a wide variety of choices for generic anomaly and threat detection</p><ul><li>Scikit-learn: Provides a wide range of algorithms for classification, regression, and clustering, useful for tasks like intrusion detection and malware classification.</li><li>TensorFlow/Keras: Powerful deep learning frameworks for building complex models, such as those used for sophisticated network intrusion detection and malware analysis.</li><li>PyTorch: Another leading deep learning framework, particularly useful for tasks involving natural language processing and computer vision within cybersecurity.</li><li>XGBoost/LightGBM/CatBoost: Gradient boosting libraries known for high accuracy in tasks like malware detection and vulnerability prediction.</li><li>Pandas: Essential for data manipulation and analysis, used to prepare cybersecurity data like logs and network traffic for model training.</li><li>NumPy: A foundational library for numerical computing, used extensively by other machine learning libraries.</li></ul><p><strong>Cybersecurity Libraries</strong></p><p>These libraries focus on specific cybersecurity tasks and can be integrated with AI/ML workflows:</p><ul><li>Scapy: Allows manipulation of network packets, useful for network intrusion detection and security testing.</li><li>Yara: A pattern-matching tool for malware analysis and detection, which can be integrated with ML models to improve detection rules.</li><li>Volatility: A memory forensics framework for analyzing RAM dumps to detect malicious activity, and can be automated with Python scripts.</li><li>Numpy-learn: Integrates Scapy with Scikit-learn for enhanced network traffic analysis using machine learning.</li></ul><p>Both AI and Cybersecurity are evolving on a daily basis and can enrich and compliment each other with time to ensure successful prevention, detection and remediation of cyberattacks thus preventing financial, customer and reputational damage to organizations.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a6ec1c32125c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Quantum Computing and AI]]></title>
            <link>https://medium.com/@umanath.nayak1/quantum-computing-and-ai-2419f300f758?source=rss-4406b526148e------2</link>
            <guid isPermaLink="false">https://medium.com/p/2419f300f758</guid>
            <dc:creator><![CDATA[Umanath Nayak]]></dc:creator>
            <pubDate>Fri, 13 Jun 2025 19:58:49 GMT</pubDate>
            <atom:updated>2025-06-14T02:22:07.373Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*btD0i4lI2H5V8q-gI0roMg.png" /></figure><p>image credit:meta.ai</p><p>IBM announced earlier this week that it had made strides of progress towards developing a large-scale, fault-tolerant quantum computer named “Quantum Starling” by 2029.</p><p><a href="https://www.upi.com/Science_News/2025/06/10/IBM-quantum-starling-computer-Arvind-Krishna/2381749586277/">IBM &#39;more certain&#39; of new quantum computer by 2029 - UPI.com</a></p><p>Also, earlier in the year , Microsoft had announced they developed a Quantum topoconductor chip Majorama-1 that will bring Quantum computing more sooner than expected, in the next couple of years.</p><p>There has been significant debate on the timeframe, use-cases and applications on Quantum based computing , the readiness and the deployments of applications bringing this form of computing to real-life.</p><p>The past two years have seen AI enter the mainstream spotlight, thanks largely to the launch of ChatGPT and other transformer-based models. These models leverage accelerated computing to tap into the immense power of large-scale GPUs. Since then, a proliferation of small and large AI models has emerged, courtesy of both open-source and closed-source providers.</p><p>Traditional computing has always relied on ‘bits’ which have a binary state of 0 or 1 . All software and computing have relied on these binary states from software development to wide ranging applications that can be generated using computers including AI based accelerated computing.</p><p>Quantum computing uses a concept called ‘qubits’ that use the rules of quantum mechanics that surpass the capabilities and speed by an exponential factor. Couple of months back, Google claimed that its quantum computing chip “Willow” solved an unsolvable math problem in 5 minutes that would take traditional computing 10 septillion years to solve(1 septillion year=10²⁴ years)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/341/1*2LcU6N7BPXcWprEkQka2Wg.jpeg" /></figure><p>While both technologies seem to be mutually exclusive, with the way computing is predicted to evolve , it is possible both will co-exist and solve completely unique use cases.</p><p><strong>AI and Quantum Computing: a closer look</strong></p><p><strong>AI:</strong></p><p>AI focuses on building intelligent systems that can learn, reason, and make decisions, often by mimicking human cognitive abilities. It relies on classical computing, but quantum computers could potentially improve AI’s performance by allowing it to process massive datasets faster and tackle more complex problems.</p><p><strong>Quantum Computing:</strong></p><p>Quantum computing utilizes qubits based quantum phenomena like superposition and entanglement to solve problems that are computationally intractable for classical computers. This includes tasks like simulating complex molecules, breaking modern encryption, and optimizing complex systems.</p><p><strong>Both can be complementary :</strong></p><p>AI and quantum computing are not directly interchangeable but rather serve as complementary tools. AI could potentially leverage quantum computing for certain aspects of its operation, such as processing large datasets or optimizing AI models, while quantum computing might benefit from AI for tasks like error mitigation and algorithm development</p><p><strong>Potential Applications:</strong></p><p>The combination of AI and quantum computing could lead to breakthroughs in various fields, including drug discovery, materials science, financial modeling, and cybersecurity.</p><p><strong>AI-Quantum Computing Intersections:</strong></p><ul><li>Quantum Machine Learning: Quantum computing can enhance Machine Learning models, such as quantum k-means or quantum support vector machines.</li><li>Quantum Neural Networks: Quantum neural networks can help build faster Neural Network models.</li><li>Quantum-inspired AI: Quantum principles can inspire new AI algorithms and techniques.</li></ul><p><strong>Algorithms:</strong></p><p><strong>AI/ML Algorithms:</strong></p><p>Depending on the use case, there are Supervised( Labeled Data) and Unsupervised (No Data) algorithms. Supervised algorithms , offered by a lot of providers use significant amount of data and compute to train and infer. Public cloud providers offer a variety of code based or non-code based solutions to develop and train these algorithmic models.</p><p>Some of the popular AI/ML models are:</p><ol><li>Linear Regression: This is one of the most predictive supervised learning statistical based models used today to predict labels(target) using dependent features( dimensions). The goal of these models are to form a linear relationship between the features to predict labels such as house/car prices, sales etc</li><li>K-Means Clustering : K-Means Clustering is one of the most popular unsupervised algorithms used to group a set of data points based on the training data and features . These algorithms help answer questions such as : What are customers based on certain traits/behaviors likely to do next? They can also be used for anomaly detection</li><li>Decision Trees: These algorithms help build a tree and relationship between the features to determine predictive and what-if scenario outcomes. Some of the popularly used examples of these algorithms are XGBoost and Random Cut Forest</li><li>Neural Networks : Neural networks are one of most complex algorithms having the ability of making decisions the same way a human brain makes decisions ( with layers of neurons). Examples of Neural Network models are CNNs( Convolutional Neural Network) , a popular model used to generate images, RNNs( Recurrent Neural Network) are used to generate text and time-series and are effective for tasks like natural language processing and speech recognition. ChatGPT is a neural network, transformer based LLM (Large Language Model).</li><li>Genetic Algorithms: These algorithms use evolution based “Survival of the fittest” sub-algorithms to help solve a problem. Based on a range of factors, a combination ( parents ) of algorithms are selected to computationally evolve and choose the best “child” to solve the problem. Use cases include scheduling tasks and finding best routes for autonomous driving.</li></ol><p><strong>Quantum Algorithms:</strong></p><p>Based on the qubits concepts explained above ( superposition and entanglement) , quantum algorithms use quantum mechanics to help solve problems exponentially faster than AI based algorithms.</p><p>Examples of Quantum Algorithms:</p><p>Shor’s Algorithm:Efficiently factors large numbers, a problem that is believed to be extremely difficult for classical computers</p><p>Grovers Algorithm: Provides a quadratic speedup for searching unsorted databases.</p><p>Hamiltonian Simulation: Allows quantum computers to simulate the behavior of quantum systems, which can be used in fields like materials science and drug discovery.</p><p>Variational Quantum Eignensolver: A quantum algorithm for solving optimization problems, including those in machine learning.</p><p>Quantum Fourier Transform(QFT): A fundamental subroutine used in many quantum algorithms, including Shor’s algorithm and Grover’s algorithm.</p><p><strong>Quantum Computing Challenges/Limitations:</strong></p><p>Quantum Error Correction:</p><ol><li>Error Rates: Quantum computers are prone to errors because of qubit and probabilistic representation</li><li>Scalability: As the number of qubits increase, the complexity increases and error rates can increase</li></ol><p>Quantum Interference:</p><ul><li>Quantum computing can be interfered because of noise sources, physical, electromagnetic interference</li></ul><p>Quantum Control and Recalibration:</p><ul><li>Maintaining precise control and calibrating qubits is essential for accurant quantum computing</li></ul><p>Quantum Hardware:</p><ol><li>Qubit quality: Improving the quality and coherence times of qubits is crucial for reliable quantum computing.</li><li>Scalability: Scaling up the number of qubits while maintaining control and low error rates is a significant challenge.</li></ol><p><strong>Quantum computing software:</strong></p><ul><li>Binary bit based software and algorithms have taken years to develop. Standard development tools , operating systems and security have evolved over time to ensure binary cryptography and security. Similar investments will be needed for quantum computing platforms.</li><li>Cloud based providers such as AWS, Google , Microsoft give their own pay-per-use quantum computing tools with simulated environments.AWS Braket, Azure Quantum and Google Quantum AI are some of these cloud provider tools. Other than public cloud there are providers such as QuTip, a free open source provider which provides Python integration</li></ul><p><strong>Some notable quantum computing researchers and experts to follow:</strong></p><ul><li>Scott Aaronson: Theoretical computer scientist and quantum computing expert.</li><li>John Preskill: Theoretical physicist and quantum computing researcher.</li><li>Michele Mosca: Quantum computing researcher and co-founder of evolutionQ.</li></ul><p>Even though quantum computing is still couple of years away, this technology will change the way computing was known in the past , will complement classical computing and has the potential to transform every industry.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2419f300f758" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>