catears124  /  ycty.ai  /  ears.cat  /  diffusion  /  ai safety  /  mechinterp  /  after effects  /  time series

catears124

diffuse

parker patton / math + cs @ ua

accelerated masters track in pure mathematics (dec. '28)

researchpilled statsmaxxer ;; welcome to my site :)

ai safety / mechinterp / diffusion / time series
python / pytorch / pandas / learning c++

price pathwaiting
L long/addS short/addC25/C100 close%TP take profitSL stop/time

motion / ae work

live from spotify

dialing spotify...

projects+experience

01
current / algoverse ai
tags: research, bilingual dataset, eval awareness, mechinterp, pytorch

Investigating unverbalized evaluation awareness in LLMs through bilingual behavioral experiments and internal-representation analysis.

  • 2,880 English/Japanese prompts
  • 6-level incentive titration
  • 2 generator models: Qwen + Gemini
  • blinded cross-model verification
02
Low-Dimensional Adversarial Attacks on Video Models
tags: research, adversarial ml, pytorch, robustness, video diffusion

Developed low-dimensional latent-space attacks that generate perceptually constrained adversarial videos for downstream classifiers.

  • 3,652 attack trials
  • 686 source videos
  • 89.5% ASR on eligible clips
  • 47% ASR from 16 latent dimensions
  • 44/90 attacks survived CogVideoX re-encoding
03
Diffusion Models for Financial Time-Series Forecasting
tags: research, diffusion, forecasting, pytorch, time series

Built a volatility-conditioned diffusion model for probabilistic financial forecasting with stronger uncertainty calibration than classical baselines.

  • 8 financial assets
  • CRPS = 0.057
  • Diebold–Mariano p = 0.002
  • MAE 0.083 vs. ARIMA 0.066
  • CARSEF Honorable Mention
04
osu!rankguess
tags: project, computer vision, data engineering, full-stack, multimodal ml

Built a deployed ML platform that predicts an osu! player's global skill percentile directly from recorded gameplay.

  • 3,000 gameplay videos
  • 1,767 unique players
  • 0.846 held-out R²
  • 1.84-point held-out MAE
  • published Kaggle dataset
05
CS2Desk
tags: project, market research, optimization, python, sql

Built a market-data and optimization engine for identifying executable positive-expectancy CS2 trade-up contracts.

  • 147,000+ Steam listings
  • 2.25M bid levels
  • 151,000+ contracts evaluated
  • 50,000 contracts searched in under 2 seconds
  • fees, slippage, liquidity, and fillability modeled
cs1shirt1shiirt2catcardRAM

resume

nameParker Patton / catears124 / ycty
schoolUniversity of Alabama / math + cs / rising sophomore / expected graduation: december 2028 (MS pure math)
looking forinternships; small teams, ownership, or learning directly under a strong engineer/researcher
locationBirmingham / Tuscaloosa / Huntsville, Alabama
workmechinterp / diffusion research / financial time series / video evals / CV / realtime interfaces
toolsPython / PyTorch / pandas / pyarrow, parquet / learning C++ / TypeScript / git, vercel, aws
currentalgoverse ai research: Researching evaluation awareness across languages under Madhur Panwar, targeting ICLR.
resumeresume.pdf

skills by evidence

PyTorch / MLdiffusion time-series model (UNet), adversarial V2V attack pipeline (probed wanVACE), osu!rankguess (catboost)
data workcustom osu! video & replay telemetry dataset collection, adversarial perturbation pipelines for K400/UCF101 classifiers
microstructureMBO datasets, FIFO backtests, CEX-DEX triangle arb, counterstrike tradeup EV tooling
systemsmonte carlo permutation tests, execution/impact simulation, learning C++
rewired_texture
contact / links

contact

updated july 2026