Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
Publications
-
Adaptive Resource Management Enabling Deception (ARMED)
ARES '17 Proceedings of the 12th International Conference on Availability, Reliability and Security
See publicationDistributed Denial of Service (DDoS) attacks routinely disrupt access to critical services. Mitigation of these attacks often relies on planned over-provisioning or elastic provisioning of resources, and third-party monitoring, analysis, and scrubbing of network traffic. While volumetric attacks which saturate a victim's network are most common, non-volumetric, low and slow, DDoS attacks can achieve their goals without requiring high traffic volume by targeting vulnerable network protocols or…
Distributed Denial of Service (DDoS) attacks routinely disrupt access to critical services. Mitigation of these attacks often relies on planned over-provisioning or elastic provisioning of resources, and third-party monitoring, analysis, and scrubbing of network traffic. While volumetric attacks which saturate a victim's network are most common, non-volumetric, low and slow, DDoS attacks can achieve their goals without requiring high traffic volume by targeting vulnerable network protocols or protocol implementations. Non-volumetric attacks, unlike their noisy counterparts, require more sophisticated detection mechanisms, and typically have only post-facto and targeted protocol/application mitigations. In this paper, we introduce our work under the Adaptive Resource Management Enabling Deception (ARMED) effort, which is developing a network-level approach to automatically mitigate sophisticated DDoS attacks through deception-focused adaptive maneuvering. We describe the concept, implementation, and initial evaluation of the ARMED Network Actors (ANAs) that facilitate transparent interception, sensing, analysis, and mounting of adaptive responses that can disrupt the adversary's decision process.
Projects
-
San Fransisco Crime Classification
-
See projectFor our Senior Design Project, my team is participating in a competition hosted by Kaggle.com.
Given a large dataset of past arrests in San Francisco, create a Machine that predicts the category of crime given the date and location.
Our program uses 3 different open source Machine Learning Libraries, over 9 classifiers and clustering techniques to accurately predict a category.
Tools used: Python, Java, WEKA, scikit-learn Java-ML
Languages
-
English
Native or bilingual proficiency
Other similar profiles
Explore more posts
-
Norman Paulsen
I've seen this a few times so I wanted to break down that math: at Claude Opus 4.6 $25 per million output tokens, this is 10 𝑩𝒊𝒍𝒍𝒊𝒐𝒏 tokens or Approximately 7.5 𝑩𝒊𝒍𝒍𝒊𝒐𝒏 words Roughly 750 𝑴𝒊𝒍𝒍𝒊𝒐𝒏 lines of code That's an estimated 15 from scratch Windows 11 rewrites Or approximately 50 new Android operating systems Google is estimated to have 2 billion lines of code across all umpteen of their products. 2 1/2 engineers at $250k in tokens could rebuild all of Google in a year. Also, AI engineer pay ranges w/bonus: Senior AI Engineer $222K–$376K Senior Machine Learning Engineer $170K–$270K Principal Machine Learning Engineer $186K–$309K Director of Machine Learning $203K–$325K So, only the top 1% of AI engineers at top tech companies are making $500k and no one person should be spending $250k on tokens unless they are rewriting all of Google solo from scratch. #AI #LLM #Nvidia #JensenHuang #SoftwareEngineering
78 Comments -
Phil Bogle
Update on new Listify features, and experiments with Wap and ChatGPT 5: Warp is running a $5 ChatGPT promotion (https://www.warp.dev/), integrated with their Agentic terminal/dev environment. I've been using Gemini Pro and the Gemini CLI so far to build Listify (https://lnkd.in/gmU5YdUE), my hobbyist AI+flutter powered list management app, but thought I'd give ChatGPT 5 a chance for comparisons sake and to see how smoothly the LLMs could hand off the codebase. Observations: - Handoff was seamless, GPT5 had no problems understanding the code base and working with it. - I really didn't see a big difference in coding capabilities between GPT 5 and Gemini Pro 2.5, but it's true this is not an extremely challenging app. - GPT5 has much better testing discipline than Gemini Pro; it suggested adding a number of tests whereas Gemini never even considered to offer. - The Warp terminal is definitely better than Gemini CLI due to the greater power that Warp offers; e.g. formatting is richer, you can upload images and easily specify files as context, and the collapsible sections provide a better experience for looking at diffs and log output. - I am still uncertain if the agentic environment of the future will look more like a browser, an IDE, or a terminal. Toggling back and forth between terminal mode and agentic mode in Warp is awkward. While you can give commands like "git push" to the LLM instead of the shell, it is terribly slow to respond. Since I last wrote, I've added the following features: - an AI-powered autosort and group feature. For example, if you have a shopping list, it will be break it into labeled sections like "produce", "dairy", "snacks", where each section contains things that are likely to be nearby in the store. This makes shopping faster. - an "Export to Google Tasks" feature to the app for people who like tahe scanning/parsing features of Listify but want to live in Google Tasks. (The backend but not UI integration was done so let GPT5 take over the UI. - more optimistic updates to make the UI feel snappier. - I also added a feature that lets you add a section header by long pressing on add item. Obigatory plug: Please LMK if you'd like to be a tester for the Android version of Listify, I still need to get to 12 testers to get it listed on the main app store for convenience of download. The web link above gives you an idea of the feature set but the mobile experience is much better. I still also need to get the iOS version up but haven't jumped through all the certifcate hoops.
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content