Memformers (gradient memory)
Treats past optimisation gradients as memory registers for procedural computation. Distinct from the 2010.06891 "Memformer" above. (Survey citation — distinct paper identifier not recovered.)
At a glance
- Type
- Past-gradient memory registers
- Tier
- T4
- Created
- 2025-08
- Latest release
- not applicable — not OSS
- License
- not applicable — not OSS
- GitHub
- not applicable — no GitHub repo
- Pricing
- not applicable — not commercial
- Funding
- not applicable — not commercial
Taxonomy
- storage
- parametric
- retrieval
- attention
- persistence
- session
- update
- parametric-edit
- unit
- weight
- governance
- opaque
- conflict
- n/a
When to use
Optimised for: not applicable - research paper
Anti-fit: not applicable - research paper
Pros & cons
Pros
Treats past optimisation gradients as memory registers for procedural computation.
Cons
Survey citation only; paper identifier not directly recovered; reproducibility limited.
Claims & capabilities
Memory-augmented transformers can implement linear first-order optimization (gradient descent, conjugate gradient, momentum); MoE adapts learned methods to OOD scenarios. Headline: competitive with classical CGD/Nesterov AGM/momentum on synthetic linear regression (no specific headline metric); baselines: Conjugate Gradient Descent, Nesterov Accelerated Gradient, Momentum Gradient Descent; primary dataset: synthetic random linear regression (d=5, n=20, Gaussian inputs).
Technical surface
- API surface
- not applicable — research paper
- Backend storage
- not applicable — research paper
- Deployment
- not applicable — not a deployable product
- Embedding model
- not applicable — research paper
- Multi-tenancy
- not applicable — research paper
- MCP
- not applicable — research paper, no deployed product
- A2A
- not applicable — research paper, no deployed product
- OpenTelemetry
- not applicable — research paper, no deployed product
Similar systems
Other recent method papers — theorized, no distinct product in the catalog, ranked by inbound references.
- Compressive Transformer T3
Maintains recent states in full resolution while compressing older memories with learned compression functions. DeepMind.
- MemGPT v2 / agent-tools T3
Already in catalog as the foundational MemGPT paper. Note: Letta is the productionised successor (cross-listed).
- Transformer-XL T3
Extends context through segment-level recurrence + caching of hidden states from prior segments. Foundational long-context architecture.
- Generative Agents T3
Park et al. — landmark agent-simulation paper. Reflection + memory stream + retrieval enable believable agent behavior.
- MemoryBank T3
Enhances LLMs with long-term memory. Early influential paper.
- Reflexion T3
Language agents with verbal reinforcement learning.
Related systems
References (21)
- A-MEM cites — S2 isInfluential citation
- AiT (Associative Transformer) cites — S2 isInfluential citation
- ARMT (Associative RMT) cites — S2 isInfluential citation
- Atlas cites — S2 isInfluential citation
- BAI-LAB MemoryOS cites — S2 isInfluential citation
- Compressive Transformer cites — S2 isInfluential citation
- Differentiable Search Index (DSI) cites — S2 isInfluential citation
- EM-LLM cites — S2 isInfluential citation
- EMAT cites — S2 isInfluential citation
- LM2 cites — S2 isInfluential citation
- MATTER cites — S2 isInfluential citation
- MemBART cites — S2 isInfluential citation
- Memformer cites — S2 isInfluential citation
- MemGPT v2 / agent-tools cites — S2 isInfluential citation
- MemLong cites — S2 isInfluential citation
- Memorizing Transformers cites — S2 isInfluential citation
- Memory Layers at Scale cites — S2 isInfluential citation
- NAMMs cites — S2 isInfluential citation
- Titans (Google) cites — S2 isInfluential citation
- Transformer-XL cites — S2 isInfluential citation
- TransformerFAM cites — S2 isInfluential citation