01. SEC_REF // COGNITIVE_MATRIX MEM_ALLOC: OK  ·  UPTIME 99.98%  ·  boot.seq nominal_
[ MACHINE LEARNING RESEARCHER ]

ETHAN
HAARER

SYS.REF // COGNITIVE-INSPIRED ARCHITECTURES

I am currently a Founding Engineer at a stealth startup where I predominately develop cognitive-inspired ML systems. Our research predominantly revolves around Representation Learning, Novel ML Algorithms, and Mechanistic Interpretability.

My research Interests lie in Neuro-Symbolic AI, specifically in cognitive-inspired systems that learn continuously from few examples, grow their own concept hierarchies the way people do, and stay explainable and interpretable at every layer.I also enjoy deploying Edge AI to see how far we can push local compute to aid those in the field.

EDU_NODE · 01
M.S. COMPUTER SCIENCE
Georgia Institute of Technology
Machine Learning Specialization
EDU_NODE · 02
B.S. COMPUTER SCIENCE
Georgia Institute of Technology
Intelligence & Devices · Summa Cum Laude
FIG.01 — LOSS SURFACE ANIMATIONR · MAPPED
02.SEC_REF
N = 003 FEATURED

Research Publications

Warped concentric gradient grid [ P·01 ]
UNSUPERVISED · LATENT_REP · EMERGENT

Deep Taxonomic Networks for Unsupervised Hierarchical Prototype Discovery

A deep latent-variable approach that optimizes a large latent hierarchy to automatically discover taxonomic structure and prototype clusters directly from unlabeled data; statistical optimization walking a gradient space toward emergent topology.

VAE MIX_OF_GAUSS NeurIPS 2025 PATENT PENDING
READ ON ARXIV
Upward hyperbolic cone wireframe [ P·02 ]
NEURO-SYMBOLIC · INTERPRETABLE · KNOWLEDGE_REP

Taxonomic Networks: A Representation for Neuro-Symbolic Pairing

Frames taxonomic networks as the shared representation that pairs connectionist learning with symbolic structure: top-down symbol instantiation and discrete knowledge routing over a learned abstraction tree that stays inspectable.

Symbolic Cobweb NEUS'25 PATENT PENDING
READ ON ARXIV
Twin-hemisphere wireframe sphere [ P·03 ]
COGNITIVE · ALIGNMENT · MULTILINGUAL

Cross-Language Typicality Effects in a Multilingual Large Language Model

Tests whether GPT-4o aligns with human typicality norms across five languages, probing the synchronization of statistical neural maps with the structured, human-like conceptual frameworks they must mirror.

GPT-4o Cognition COGSCI'25
LINK TO RESEARCH
[03.SEC_REF]// CAREER TIMELINE
N = 003 · deployment log

Work Experience

[INDEX]
[ORG]
[IDENTIFIER / DESCRIPTION]
[TIMEFRAME]
{{ r.idx }}
{{ r.title }}
{{ r.desc }}
{{ r.meta }}
{{ r.time }}
[03B.SEC_REF]// EXPANSIBLE_REPOSITORY
N = 008 · CLICK ROW TO TOGGLE [ ± ]

Projects Showcase

{{ r.title }}
FIG · MESH_XZ DRAG · ORBIT
FIG · FROZEN_FRAME DRAG · ORBIT
{{ r.figA }}● LIVE_LOOP
{{ r.figB }}R · MAPPED
{{ r.chartAlt }}
TITLE

{{ r.title }}

STATUS
{{ r.status }}
CLASS_REF
{{ r.classRef }}
STACK
{{ r.meta }}
DESCRIPTION

{{ r.desc }}

ACTIONS
04.SEC_REF // ABSTRACTION_TERMINAL
awaiting operator input
Square gravity-sink wormhole

OPEN A
CHANNEL

DATA SINK · INBOUND QUERIES
SEND PACKET
// DATA_NODES · EXTERNAL_INDICATORS
[OK] channel encrypted · handshake nominal
[R] RESTRICTED · operator response < 48h