More about HKUST
Federated Knowledge Transfer and Co-Optimization of Large and Small Language Models
PhD Thesis Proposal Defence
Title: "Federated Knowledge Transfer and Co-Optimization of Large and Small
Language Models"
by
Mr. Tao FAN
Abstract:
Large Language Models (LLMs) have achieved remarkable success in language
understanding, generation, and reasoning. However, their practical deployment
in domain-specific and privacy-sensitive environments remains difficult. In
many real-world scenarios, valuable knowledge is distributed across
institutions, enterprises, and edge devices, where clients possess private
domain data but cannot directly share them with a central LLM server.
Meanwhile, these clients often rely on efficient Small Language Models (SLMs)
because full-scale LLM deployment and fine-tuning are prohibitively expensive.
This creates a fundamental challenge: how to enable effective knowledge
transfer and collaborative improvement between powerful server-side LLMs and
efficient client-side SLMs under constraints of privacy, model ownership,
resource limitation, and heterogeneity.
This thesis proposal addresses this challenge through a unified perspective of
federated knowledge transfer and co-optimization of large and small language
models. We establish the foundations and taxonomy of federated large-small
language models by organizing the problem space along three complementary
paradigms: downward transfer from LLMs to SLMs, upward transfer from SLMs to
LLMs, and bidirectional co-optimization between the two. We further categorize
transfer mechanisms across data-level, representation-level, and model-level
interfaces, clarifying the trade-offs among utility, efficiency, privacy, and
model intellectual-property protection.
Building on this foundation, the proposal presents three completed frameworks.
First, we propose FedCoT, a privacy-preserving framework for transferring
reasoning knowledge from LLMs to SLMs. FedCoT perturbs client prompts with
Differential Privacy and exploits Chain-of-Thought (CoT) rationales generated
by the server-side LLM to improve the reasoning ability of client SLMs without
exposing sensitive queries.
Second, we introduce PPC-GPT, a federated task-specific compression framework
that compresses LLMs into lightweight SLMs through rationale-aware structured
pruning and CoT-guided synthetic data generation, thereby preserving critical
reasoning capability under resource constraints.
Third, we develop FedMKT, a framework for bidirectional co-optimization of
LLMs and SLMs. FedMKT performs selective mutual knowledge distillation and
resolves architectural and tokenizer mismatch through Minimum Edit Distance
(MinED) token alignment, enabling both server and client models to benefit
from each other.
Extensive experiments on diverse text generation and reasoning benchmarks show
that the completed methods consistently outperform existing federated
baselines. For future work, we will study FedProxy for heterogeneity-aware
upward knowledge fusion from client-side SLMs to a proprietary server-side
LLM. FedProxy aims to address the trilemma of LLM intellectual-property
protection, client data privacy, and reliable knowledge fusion by using
high-fidelity Proxy SLMs, heterogeneity-aware aggregation, and training-free
plug-in fusion. Overall, this proposal establishes a principled and practical
foundation for privacy-aware federated knowledge transfer between large and
small language models, and outlines a concrete path toward collaborative,
efficient, and secure deployment of language intelligence in distributed
environments.
Date: Monday, 21 September 2026
Time: 2:00pm - 4:00pm
Venue: Room 5501
Lift 25/26
Committee Members: Prof. Qiang Yang (Supervisor)
Prof. Kai Chen (Co-supervisor)
Dr. Dan Xu (Chairperson)
Dr. May Fung