M.Phil. researcher · HKUST(GZ)

Open to research & algorithm internships

I study how language modelsrepresent, transport, and revise reasoning.

I build causal tests and reliable interventions for latent reasoning, retrieval, and model evaluation.

EMNLP 2026First-author paper
EMNLP 2026FCPRAG
MICCAI 2024Best Method · 3rd place

Selected work

Each project is presented with its question, evidence, and boundary—not only its headline result.

Research that connects mechanism, intervention, and evaluation.

01EMNLP 2026 · First author

SCIT: Testing Causal Cache Carriers in Latent Chain-of-Thought Models

A causal audit for identifying cache objects that carry counterfactual computation.

Latent CoTCausal interventionKV cache
View research noteClose note
Question
Which cached objects actually carry a model’s counterfactual computation?
Approach
Source–recipient counterfactuals, K/V component replacement, semantic controls, and matched corruption.
Evidence
A competence-gated carrier map across latent-CoT models, with value-cache suffix trajectories prominent in the examined GPT-2 cells.
Boundary
Cells that fail the base-task competence gate receive no mechanism claim.
02ICLR 2027 · In preparation

Transport vs. Destructive Sensitivity

Separating answer transport from destructive sensitivity in causal-cache interventions.

Causal cacheRelayReplication
View research noteClose note
Question
When an intervention changes an answer, did it transport computation or merely damage it?
Approach
A two-estimand framework with crossed overwrites and independently trained checkpoints.
Evidence
Prospective transport patterns and an isolated three-token relay, with directional replication on CODI-Llama-1B.
Boundary
Transportability is not treated as proof of the model’s natural computation path.
03EMNLP 2026 · Accepted

FCPRAG: Fusion-Controller Parametric Retrieval-Augmented Generation

A lightweight controller for stable multi-passage LoRA injection under retrieval uncertainty.

RAGLoRA fusionUncertainty
View research noteClose note
Question
How can parametric RAG combine passage adapters without brittle, uniform fusion?
Approach
Predicts fusion weights, gates, and temperatures, with a conservative fallback under uncertainty.
Evidence
Evaluated on four QA datasets and three LLM backbones; reported gains reached 4.65% on 2WikiMultiHopQA and 7.55% on CWQ.
Boundary
Gains are reported for the evaluated retrieval and backbone settings, not as universal improvement.
04NeurIPS 2026 · Submitted

Target Selection Margin for Causal SAE Feature Selection

Target-specific SAE feature selection evaluated through directional interventions.

Sparse autoencodersInterventionFeature selection
View research noteClose note
Question
Does a selected feature transfer from teacher-forced scoring to held-out interventions?
Approach
A target–distractor probability margin with label-shuffle controls and cross-model ablations.
Evidence
Compared target-aware, random, and generic utility selectors across Qwen, Gemma, and Llama.
Boundary
Teacher-forced margins are reported separately from visible generation changes.
05ICML 2026 · Submitted

REWIND: Improving LLM Reasoning Quality at Inference Time

An inference-time controller that locates risky spans, backtracks, and selectively retains better trajectories.

Inference-time controlReasoningEvaluation
View research noteClose note
Question
Can a model repair a reasoning trajectory without restarting the whole response?
Approach
Entropy-gradient localization, semantic-boundary backtracking, suffix regeneration, and MH-style acceptance.
Evidence
Evaluated across five reasoning benchmarks and three model families, with the clearest gains on difficult mathematics.
Boundary
Quality gains are considered together with additional inference cost.
06MICCAI 2024 · Sole author

MARIO Challenge: Medical Image Prediction

OCT sequence diagnosis and progression prediction for neovascular AMD.

Medical imagingOCTRobust ML
View research noteClose note
Question
How can temporal OCT changes support reliable diagnosis and patient-level risk prediction?
Approach
ConvNeXt V2 features, Siamese OCT-DiffNet, CLAM-SB multiple-instance learning, and weighted sampling.
Evidence
Two sole-authored LNCS papers, third place, Best Method, and an F1 score above 0.84.
Boundary
Challenge results are presented as application-specific evidence, not clinical validation.

Publications & manuscripts

Work, separated by evidence status.

Accepted & published

2026EMNLP

SCIT: Testing Causal Cache Carriers in Latent Chain-of-Thought Models

Yi Ding, Lijun Huang, Menglin Yang

First author · Accepted
2026EMNLP

FCPRAG: Fusion-Controller Parametric Retrieval-Augmented Generation for Stable Multi-Passage LoRA Injection

Jinchang Zhu, Jindong Li, Yi Ding, Xiaojian Nie, Rong Fu, Shuangyong Song, Haowei He, Menglin Yang

Accepted
2024MICCAI / LNCS 15503

Two contributions in Image-Driven Prediction of Retinal Disease Progression

Yi Ding

Sole author · Best Method

Manuscripts

2026ICLR 2027

SCIT Extension: Transport vs. Destructive Sensitivity

Yi Ding, Menglin Yang

In preparation
2026NeurIPS

Target Selection Margin for Causal SAE Feature Selection

Yi Ding, Menglin Yang

Submitted
2026ICML

REWIND: Improving LLM Reasoning Quality

Yi Ding, Menglin Yang

Submitted

Curriculum vitae

Research training and technical range.

Education

2025—2027
Hong Kong University of Science and Technology (Guangzhou)

M.Phil. in Artificial Intelligence

2024—2025
University of Edinburgh

M.Sc. in Artificial Intelligence

2020—2024
University of Nottingham

B.Sc. (Hons) in Artificial Intelligence · First Class

Experience

2024
Research Assistant · Shenzhen Bay Laboratory

Protein representation, uncertainty-aware modeling, and reproducible experimental reporting.

Capabilities

LLM systems

Latent reasoning · RAG · Agents · Tool use

Training & evaluation

PyTorch · Transformers · LoRA · PPO/RL · SFT

Research engineering

Python · Git · Linux/HPC · SLURM · Reproducible pipelines

About

A research engineer’s view of reliable AI.

I am an M.Phil. student in Artificial Intelligence at HKUST(GZ). My work sits between mechanistic experimentation and practical system building: I design controlled interventions, audit what the evidence supports, and turn research workflows into reproducible engineering artifacts.

Hong Kong · GuangzhouLLM reasoning · Reliable AI

Interested in reasoning systems, causal evaluation, or research engineering?

Start a conversation