HumemAI iconHumemAI

Projects / Machines With Human-Like Memory

Subproject

RoomKG Baselines

The RoomKG benchmark and baseline implementations accompanying Temporal Knowledge-Graph Memory in a Partially Observable Environment.

Benchmark

Reframe the memory problem in temporal knowledge-graph terms.

Figure 1

Temporal knowledge-graph framing

  • Temporal knowledge-graph memory
  • Partial observability retained
  • Explicit benchmark framing
  • Shared evaluation surface
The benchmark shifts the environment and memory question into a graph-native formulation.

RoomKG Baselines turns the earlier room-memory benchmark into a more explicit graph-memory setting. Observations, hidden state, and memory can now be represented and studied as temporal knowledge graphs rather than only as symbolic records or latent policy state.

That gives the research line a stronger benchmark object for comparing graph-based memory approaches under partial observability.

Baselines

Compare symbolic and neural approaches under one setup.

Figure 2

Baseline comparison layer

  • Symbolic baselines
  • Neural baselines
  • Shared evaluation protocol
  • Reusable benchmark artifacts
The repository packages multiple baseline styles around one benchmark instead of one isolated method.

The repository collects baseline implementations so symbolic, neural, and hybrid approaches can be evaluated under the same benchmark conditions. That makes the work more than a paper artifact: it is the comparative layer around the benchmark itself.

Instead of burying the comparison inside one system, the project makes the baseline surface explicit and reusable.

Direction

Push the broader project toward graph-based long-term memory.

Figure 3

Later-phase research direction

  • Later stage in the project line
  • Graph-structured memory focus
  • Benchmark plus baselines
  • Foundation for follow-on work
The project moves the research line toward graph-structured memory and stronger benchmark comparison.

RoomKG Baselines marks a later phase in Machines With Human-Like Memory, where the emphasis shifts from explicit memory systems in general to temporal knowledge-graph memory in particular.

It is the point where the benchmark, the implementation surface, and the research question all align around graph-structured memory.

Resources

Paper and code.