Enhanced Mycelium of Thought (EMoT): A Bio-Inspired Hierarchical Reasoning Architecture with Strategic Dormancy and Mnemonic Encoding

Florian Odi Stummer

cs.AIMar 25, 2026

Enhanced Mycelium of Thought (EMoT): A Bio-Inspired Hierarchical Reasoning Architecture with Strategic Dormancy and Mnemonic Encoding

Authors: Florian Odi Stummer

Organizations: Institute of General Medicine Martin Luther University Halle-Wittenberg, Germany

Abstract

Current prompting paradigms for large language models (LLMs), including Chain-of-Thought (CoT) and Tree-of-Thoughts (ToT), follow linear or tree-structured reasoning paths that lack persistent memory, strategic dormancy, and cross-domain synthesis. We present the Enhanced Mycelium of Thought (EMoT) framework, a bio-inspired reasoning architecture that organises cognitive processing into a four-level hierarchy (Micro, Meso, Macro, Meta), implements strategic dormancy and reactivation of reasoning nodes, and integrates a Memory Palace with five mnemonic encoding styles. EMoT is a research prototype for complex, multi-domain problems, not a general-purpose prompting enhancement. Two complementary evaluations reveal a characteristic trade-off. In a blind LLM-as-Judge evaluation across three domains, EMoT achieved near-parity with CoT (4.20 vs. 4.33/5.0) with higher stability, and outperformed CoT on Cross-Domain Synthesis (4.8 vs. 4.4). Ablation studies show that strategic dormancy is architecturally essential (quality collapsed from 4.2 to 1.0 when disabled). On a 15-item short-answer benchmark, EMoT (27%) substantially underperformed simpler baselines, confirming systematic overthinking on simple problems. These results are subject to important limitations: small sample sizes (n=3 complex cases, n=15 short-answer items), LLM-as-Judge evaluation with potential self-preference bias, and approximately 33-fold computational cost overhead. To our knowledge, EMoT is the first reasoning framework to combine hierarchical topology, strategic thought dormancy with reactivation, and mnemonic memory encoding in a single architecture.

Figures & tables

Property	CoT	ToT	GoT	Reflexion	EMoT
Topology	Linear	Tree	Graph	Linear + Memory	Hierarch. network
Parallel paths	No	Yes	Yes	No	Yes
Thought preserv.	No	No (pruned)	No	Verbal summary	Dormant nodes
Reactivation	No	No	No	Partial	Yes (ctx-sens.)
Persistent mem.	No	No	No	Verbal buffer	Mem. Palace (5 enc.)
Cross-dom. synth.	No	Limited	Possible	No	Explicit (IDE)

Table 1: Comparison of reasoning framework properties.

Run	EMoT Score	CoT Score	Winner
Run 1	4.2	4.2	Tie
Run 2	4.2	4.5	CoT
Run 3	4.2	4.3	CoT
Mean (SD)	4.20 (0.00)	4.33 (0.15)	CoT

Table 2: Per-run overall scores: blind-judge quality benchmark.

Criterion	EMoT Mean	CoT Mean	Winner
Recursion Depth	4.0	4.6	CoT
Dormant Thought Management	2.9	3.1	CoT
Cross-Domain Synthesis	4.8	4.4	EMoT
Memory Utilisation	4.0	4.2	CoT
Structured Output	5.0	5.0	Tie
Solution Quality	4.6	4.7	CoT

Table 3: Per-criterion means across three blind evaluation runs (mean scores, 1–5 scale).

Figure 1: Radar chart comparing EMoT and CoT across six evaluation criteria. EMoT outperforms CoT on Cross-Domain Synthesis (4.8 vs. 4.4) while CoT leads on Recursion Depth and Solution Quality. Both achieve parity on Structured Output.

Figure 2: Run-to-run stability comparison. EMoT produces identical overall scores across all three independent runs (SD=0.00), while CoT exhibits greater variability (SD=0.15).

Configuration	Overall Score	Delta from Full EMoT
Full EMoT	4.20	–
No Dormancy ( --no-dormancy )	1.00	$-3.20$
No Memory Palace ( --no-memory-palace )	4.10	$-0.10$

Table 4: Ablation study results: overall quality scores (1–5 scale).

Figure 3: Ablation study results. Disabling the Strategic Dormancy Controller causes a catastrophic quality drop from 4.20 to 1.00 ( $-3.20$ ), while removing the Memory Palace produces a modest decrease to 4.10 ( $-0.10$ ).

Criterion	EMoT	CoT	Winner
Recursion Depth	4	4	Tie
Dormant Thought Management	3	2	EMoT
Cross-Domain Synthesis	4	3	EMoT
Memory Utilisation	4	4	Tie
Structured Output	5	5	Tie
Solution Quality	4	4	Tie

Table 5: Per-case scores for Patient Bengt (blind evaluation, Run 1).

Model	Tier	Avg Score	Runtime	Tokens
Claude Sonnet 4	Advanced	3.50	176 s	19,438
Stub (baseline)	Baseline	1.00	0.1 s	0

Table 6: Cross-model sanity check results.

Technique	Math	Logic	M-QA	Plan.	BBH	Total
Direct	3/3	3/3	3/3	3/3	3/3	15/15 (100%)
CoT	3/3	1/3	3/3	2/3	2/3	11/15 (73%)
Self-Cons.	3/3	1/3	3/3	1/3	1/3	9/15 (60%)
EMoT	1/3	1/3	1/3	1/3	0/3	4/15 (27%)

Table 7: Multi-technique accuracy benchmark: correct answers per category (3 problems each, 15 total).

Technique	Avg Tokens	Avg Time	Cost (15 problems)
Direct	81	1.8s	$0.01
CoT	414	6.4s	$0.06
Self-Consistency	1,236	26.9s	$0.17
EMoT	12,136	183.4s	$1.64

Table 8: Efficiency comparison across techniques (averaged over 15 problems).

Figure 4: Trade-off between short-answer accuracy (multi-technique benchmark) and complex reasoning quality (blind-judge benchmark). EMoT underperforms on factual accuracy but achieves near-parity with CoT on complex multi-domain reasoning quality.

Metric	EMoT	CoT	Ratio
LLM calls	99	3	33x
Tokens (EMoT engine)	${\sim}79{,}052$	${\sim}3{,}000$ (est.)	${\sim}26$ x
Runtime	${\sim}1{,}214$ s	${\sim}97$ s	${\sim}13$ x
Cost estimate	${\sim}\$ 0.36$	${\sim}\$ 0.02$ (est.)	${\sim}18$ x

Table 9: Computational cost comparison (3 test cases).

Figure 5: Per-case, per-run Cross-Domain Synthesis scores. EMoT’s advantage is most consistent on the clinical reasoning case (Patient Bengt), where cross-domain integration is most demanding.

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Case	Criterion	EMoT	CoT
Bengt	Recursion Depth	4	4
Bengt	Dormant Thought	3	2
Bengt	Cross-Domain Synthesis	4	3
Bengt	Memory Utilisation	4	4
Bengt	Structured Output	5	5
Bengt	Solution Quality	4	4

Appendix

Table 15

Case	Criterion	EMoT	CoT
Bengt	Recursion Depth	4	4
Bengt	Dormant Thought	3	3
Bengt	Cross-Domain Synthesis	4	4
Bengt	Memory Utilisation	4	4
Bengt	Structured Output	5	5
Bengt	Solution Quality	4	4

Appendix

Table 16

Case	Criterion	EMoT	CoT
Bengt	Recursion Depth	4	4
Bengt	Dormant Thought	3	2
Bengt	Cross-Domain Synthesis	5	3
Bengt	Memory Utilisation	4	4
Bengt	Structured Output	5	5
Bengt	Solution Quality	5	4

Appendix

Table 17

Criterion	EMoT Mean	CoT Mean	Winner
Recursion Depth	4.0	4.6	CoT
Dormant Thought Management	2.9	3.1	CoT
Cross-Domain Synthesis	4.8	4.4	EMoT
Memory Utilisation	4.0	4.2	CoT
Structured Output	5.0	5.0	Tie
Solution Quality	4.6	4.7	CoT

Appendix

Table 18

Configuration	Overall Score
Full EMoT	4.20
No Dormancy ( --no-dormancy )	1.00
No Memory Palace ( --no-memory-palace )	4.10

Appendix

Table 19

Criterion	EMoT	CoT	Winner
Recursion Depth	4	4	Tie
Dormant Thought Management	3	3	Tie
Cross-Domain Synthesis	5	5	Tie
Memory Utilisation	4	4	Tie
Structured Output	5	5	Tie
Solution Quality	4	5	CoT

Appendix

Table 20

Figure 6: Representative reasoning trace for the Patient Bengt case, illustrating EMoT’s hierarchical processing across four levels. Dormant nodes (dashed borders) are preserved across iterations and reactivated when new evidence warrants reconsideration, demonstrating the Strategic Dormancy Controller’s role in maintaining alternative hypotheses.

LLM Prompting

Large language model prompting studies how instructions, examples, context, and formats steer model behavior at inference time.

See all 429 papers

Hierarchical Chain-of-Thought: Enhancing LLM Reasoning Performance and Efficiency
Mar 31, 2026Xingshuai Huang, Derek Li, Bahareh Nikpour +1CoT Reasoning LLM Reasoning
Huawei Technologies Canada, Canada · Huawei Noah’s Ark Lab, Canada
Structured Thoughts For Improved Reasoning And Context Pruning
Jul 11, 2026Zain Sarwar, Supriyo Chakraborty, Berkcan Kapusuzoglu +5Efficient Language Model Reasoning Structured CoT Reasoning
University of Chicago · Capital One
Heuristic Classification of Thoughts Prompting (HCoT): Integrating Expert System Heuristics for Structured Reasoning into Large Language Models
Apr 14, 2026Lei Lin, Jizhao Zhu, Yong Liu +3Structured Reasoning in Language Models Constraint-Guided Reasoning in Language Models
Computer Network Information Center, Chinese Academy of Sciences, Beijing, China · Shenyang Aerospace University · ZGC LAB, Beijing,China +1

Mar 31, 2026J/K move · Enter open · S save

Hierarchical Chain-of-Thought: Enhancing LLM Reasoning Performance and Efficiency

Xingshuai Huang, Derek Li, Bahareh Nikpour, Parsa Omidi

Huawei Technologies Canada, Canada · Huawei Noah’s Ark Lab, Canada

CoT Reasoning LLM Reasoning Efficient Language Model Reasoning

Open paper AskPDF arXiv

Chain-of-Thought (CoT) prompting has significantly improved the reasoning capabilities of large language models (LLMs). However, conventional CoT often relies on unstructured, flat reasoning chains that suffer from redundancy and suboptimal performance. In this work, we introduce Hierarchical Chain-of-Thought (Hi-CoT), a structured reasoning paradigm specifically designed to address the challenges of complex, multi-step reasoning. Hi-CoT decomposes the reasoning process into hierarchical substeps by alternating between instructional planning and step-by-step execution. This decomposition enables LLMs to better manage long reasoning horizons and maintain logical coherence. Extensive evaluations across diverse LLMs and mathematical reasoning benchmarks show that Hi-CoT consistently improves average accuracy by 6.2% (up to 61.4% on certain models and tasks) while reducing reasoning trace length by 13.9% compared to CoT. We further show that accuracy and efficiency are maximized when models strictly adhere to the hierarchical structure. Our code is available at https://github.com/XingshuaiHuang/Hi-CoT.

Abstract

LLM Prompting

Explore similar work

Hierarchical Chain-of-Thought: Enhancing LLM Reasoning Performance and Efficiency

Structured Thoughts For Improved Reasoning And Context Pruning

Heuristic Classification of Thoughts Prompting (HCoT): Integrating Expert System Heuristics for Structured Reasoning into Large Language Models