Mole Project Permalink
Published:
Created a robotic mole plush toy that talks and takes voice commands for mole day.
Posts by Collection
projects
Blind Assist Permalink
Published:
Created a headset and program to be used by visually impaired pedestrians.
Covid Detection Kaggle Permalink
Published:
Created a model to view lung X-rays to localize lung opacities and use them to make COVID-19 diagnoses.
Vision Transformer Interpretability Permalink
Published:
Interpreted a vision transformer. I made changes to the lucent library, created a frontend to visualize things, and made a slideshow write-up of my work.
Unsupervised Disentanglement Permalink
Published:
Conducted a literature review of unsupervised disentanglement and ran experiments with disentanglement and scale. Here are the report and math notes.
Charizzma Permalink
Published:
Charizzma is an AI conversational advice-giver that helps you respond smoothly when you need it most. Specifically, we used real-time transcription, keyword extraction, and speech synthesis to deliver advice in real time.
LangShell Permalink
Published:
Our shell tool, called LangShell, is a GPT-3 based shell assistant with a memory. You can tell LangShell facts about yourself, and you will never forget them (unless you want it to). We built LangShell using Python, the OpenAI GPT-3 API, and the Huggingface MiniLM-L6-v sentence similarity API. It is based on a fork of Shell GPT, and we plan to release it as a Pull request on the original repo.
LLM Trajectory Analysis Permalink
Published:
Analyzed trajectories of language model embeddings projected onto a 3-sphere using UMAP. Won the special Nomic bounty at the NYU Generative AI Betaworks Hackathon.
Decoding Behavior from Neural Data Permalink
Published:
Designed new and implemented existing algorithms for decoding mouse wheel speed behavior from neuron-level brain recordings.
publications
Machine Learning and Proactive Network Maintenance: Transforming Today’s Plant Operations
Published in Fall Technical Forum: SCTE, NCTA, CableLabs, 2021
This paper discusses uses machine learning to classify types radio frequency impairments in cable modems.
Recommended citation: Berkan Ottlik, Brady Volpe. (2021). "Machine Learning and Proactive Network Maintenance: Transforming Today's Plant Operations" 2021 Fall Technical Forum: SCTE, NCTA, CableLabs. https://www.nctatechnicalpapers.com/Paper/2021/2021-machine-learning-and-proactive-network-maintenance-transforming-today-s-plant-operations/
The Effect of Model Capacity on the Emergence of In-Context Learning
Published in ICLR 2024 Workshop on Understanding of Foundation Models (ME-FoMo), 2024
This paper investigates the relationship between model capacity and the emergence of in-context learning under a simplified statistical framework in the transformer model.
Recommended citation: Berkan Ottlik, Narutatsu Ri, Daniel Hsu, Clayton Sanford. (2024). "The Effect of Model Capacity on the Emergence of In-Context Learning" ICLR 2024 Workshop on Understanding of Foundation Models (ME-FoMo). https://openreview.net/pdf?id=YZM9g0Mi9a
A Sequential Lightbulb Problem
Published in Presented at the Fall Fourier Talks, University of Maryland, 2024
This paper studies an online version of the light bulb problem and gives simple lower-bounds and develops algorithms that allude to a space-runtime tradeoff.
Recommended citation: Noah Bergam, Berkan Ottlik, Arman Özcan. (2024). "A Sequential Lightbulb Problem". https://berkan.xyz/files/lightbulb.pdf
Gradient Flow Dynamics of Teacher-Student Distillation with the Squared Loss
Published in Presented at the Summer@Simons poster session., 2024
This paper studies the gradient flow dynamics teacher-student distillation setup with squared loss and multiple student and teacher neurons.
Recommended citation: Berkan Ottlik. (2024). "Gradient Flow Dynamics of Teacher-Student Distillation with the Squared Loss". https://berkan.xyz/files/underparameterizedDynamics.pdf