https://www.brissettecj.com/research daily 0.75 2023-08-29 https://www.brissettecj.com/research/coarsening monthly 0.5 2023-09-13 https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/7196d0e0-e6c3-4b83-8fe6-57327fec6266/Frame+3.png Research - Graph Coarsening. - Make it stand out Figure 1: A visualization of a multi level method (multigrid [6]) for solving Laplace’s equation on a simple grid with Dirichlet boundary conditions on the top and bottom. Here blue is cold and red is hot. In this case there are only two levels of coarsening. https://www.brissettecj.com/research/gnn monthly 0.5 2023-09-13 https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/0f28a7ac-ce09-4c00-9f5d-cc6a8a0e8b28/Frame+19+%281%29.png Research - Graph Neural Networks. - Make it stand out Figure 2: A visualization of the averaging done in a GCN. In this case the nodes are simply averaged with their neighbors and the result for each entry is truncated to the first decimal place. In an actual Kipf and Welling GCN the averaging operation is more complicated, but this gets the idea across! The color of each node is given by its vector embedding taken as an RGB value associated with each entry. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/6f3cbe04-b25a-45a3-9283-efd0245f9f21/snippy.PNG Research - Graph Neural Networks. - Make it stand out Figure 3: Accuracy and loss curves for GCN training using Koopman accelerated Adam (red-solid) versus standard Adam (black-dashed). These are all performed on the Cora data set, and from left to right, have the maximum learning rates of 0.0001, 0.001, and 0.01 respectively. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/eac27537-46f7-466d-924c-bec9f71f2bfb/Frame+11.png Research - Graph Neural Networks. - Make it stand out Figure 1: A visualization of node classification using graph embedding. Nodes are given embeddings (shown as the colored bars/vectors above each node on the left), and some classification is performed on top of those classification which yields classes shown on the right by the nodes color. The task of node-embedding is obtaining these colored vectors shown on the left. https://www.brissettecj.com/research/nullmodels monthly 0.5 2023-09-13 https://www.brissettecj.com/research/networked-dynamics monthly 0.5 2023-08-28 https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/ff070ac1-9c6f-4016-b716-0ab2c8fb068f/Screenshot+from+2023-08-28+10-40-45.png Research - Networked Dynamics - Make it stand out Figure 2: Speed-ups provided by increasing the number of ranks on a test suite of real world graphs. Each run here is an initial mean-field approximation plus ten rounds of message passing. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/31ac749c-0eea-445e-b1d8-a448694ed269/Screenshot+from+2023-08-28+10-43-20.png Research - Networked Dynamics - Make it stand out Figure 1: The proportional steady-state error obtained by the mean-field approximations of gene-regulatory dynamics on Erdos Renyi and Barabasi Albert models respectively. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/b9709173-c1a5-4b25-a4d5-832e003e4466/Screenshot+from+2023-08-28+11-27-02.png Research - Networked Dynamics - Make it stand out Figure 3: Control cost and total cost for controlling the global risk network using LQR control. The average cost for a collection of 7 randomly chosen nodes is shown as the green bars, and the average cost is represented by a grey dotted line. In the paper, a collection of 7 specifically chosen nodes is compared against, and is represented by the red dotted line. https://www.brissettecj.com/tech-blog daily 0.75 2023-11-18 https://www.brissettecj.com/tech-blog/detecting-a-lying-language-model monthly 0.5 2023-11-18 https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/58488dec-6e11-4e07-a735-9d39ec3d0032/Secondary_Check.PNG Technical-Blog - Detecting A Lying Language Model. - Make it stand out Whatever it is, the way you tell your story online can make all the difference. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/ee344691-0c5a-432c-8f90-aee751c314cf/GEICO-game-rules.PNG Technical-Blog - Detecting A Lying Language Model. - Make it stand out Whatever it is, the way you tell your story online can make all the difference. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/53cba422-db8f-45ce-9937-8ea36f0aabdc/LLMs_stacked.png Technical-Blog - Detecting A Lying Language Model. - Make it stand out Figure 1: An image of multiple different attack possibilities. Here, red means compromised, and green means uncompromised. The blobs at the top represent people, the boxes represent individual instances of our LLM, and arrows represent prompts. The man-in-the-middle is red, and our regular user is green. The order of events within each box transpires alphabetically. Box 1: The standard communication between user and service. In this case the user sends an uncompromised prompt to the LLM, and receives an uncompromised response. Box 2: This is the basic man-in-the-middle case. In this case a nefarious user sends a compromised message to the LLL which changes its behavior before the regular user can use the service. This user then receives a compromised answer to their prompt! Box 3: This is the “double LLM” strategy. In this case, there is an LLM which the user cannot communicate with that checks the output of the first LLM for any compromised outputs. Then if it detects one, it either ends the communication, or informs the most recent user of the security risk. Box 4: This represents a potential problem with the double LLM strategy. In this case, the MITM sent a prompt which not only compromised the first LLM, but caused that LLM to output something which compromised the quarantined model as well. This then becomes just a more convoluted version of Box 2. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/904afd32-dc9e-4b5e-9a0d-ccf15ca2a7bf/NewAttack.PNG Technical-Blog - Detecting A Lying Language Model. - Make it stand out Whatever it is, the way you tell your story online can make all the difference. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/a1b70f07-ec05-42bb-8f24-56213752cb19/The4thwall.PNG Technical-Blog - Detecting A Lying Language Model. - Make it stand out Whatever it is, the way you tell your story online can make all the difference. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/72dfa8ac-36de-48f8-9173-b5be6ac8233a/Progressive.PNG Technical-Blog - Detecting A Lying Language Model. - Make it stand out Whatever it is, the way you tell your story online can make all the difference. https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/29910220-fb28-4638-8995-1e99f28e94a5/GEICO-water-dam.PNG Technical-Blog - Detecting A Lying Language Model. - Make it stand out Whatever it is, the way you tell your story online can make all the difference. https://www.brissettecj.com/tech-blog/multilevel-schemes-without-memory-movement-on-gpu monthly 0.5 2023-09-13 https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/3edc645b-8bd8-4494-8487-4861134ef5a2/Frame+13.png Technical-Blog - Multilevel Schemes Without Memory Movement. - Make it stand out Whatever it is, the way you tell your story online can make all the difference. https://www.brissettecj.com/tech-blog/the-shaky-grounds-of-modularity-maximization monthly 0.5 2023-08-29 https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/3f4d5ec2-fa20-443e-a9f0-031405b24d6d/Frame+10.png Technical-Blog - The Shaky Grounds of Modularity Maximization. - Make it stand out Whatever it is, the way you tell your story online can make all the difference. https://www.brissettecj.com/home daily 1.0 2024-09-11 https://images.squarespace-cdn.com/content/v1/649dc97cce656d1110535627/16867343-b470-437c-9feb-673e31381258/Screenshot_20230623-144921_Gallery+%281%29.jpg https://www.brissettecj.com/about daily 0.75 2023-07-30 https://www.brissettecj.com/contact daily 0.75 2023-07-30 https://www.brissettecj.com/pictures daily 0.75 2023-07-30