Sherlock: A Framework for Inference Attention

chevron-icon
RFP Proposals
Top
chevron-icon
project-presentation-img
Expert Rating 2.6
Luke Mahoney (MLabs)
Project Owner

Sherlock: A Framework for Inference Attention

Expert Rating

2.6

Overview

OpenCog’s Hyperon framework offers a way to store and process knowledge, a foundational requirement for AGI. In order for this knowledge to be useful, however, it must be retrievable when most relevant, in a process known as Attention Allocation (AA). Economic attention networks (ECANs) provided Attention Allocation for OpenCog classic. While they offer a technique for identifying useful information, they currently lack the ability to always employ the best system for subsequently processing that information.

RFP Guidelines

Framework for evaluating approaches to attention allocation

Complete & Awarded
  • Type SingularityNET RFP
  • Total RFP Funding $60,000 USD
  • Proposals 5
  • Awarded Projects 2
author-img
SingularityNET
Oct. 9, 2024

The goal of this project is to develop a framework to evaluate various approaches to Attention Allocation (AA) within the OpenCog Hyperon and PRIMUS architectures. The AA system dynamically allocates cognitive resources to Atoms in the Distributed Atomspace (DAS), and this framework will help assess AA methods based on desired cognitive dynamics. The framework will improve both Probabilistic Logic Networks (PLN) and evolutionary methods like Meta-Optimizing Semantic Evolutionary Search (MOSES), which are critical components of the PRIMUS architecture.

Proposal Description

Company Name (if applicable)

MLabs LTD

Project details

We believe that this is a key shortcoming of ECANs. It is not sufficient to highlight atoms of knowledge that are relevant: we must also decide how best to synthesize these data, whether it be using Probabilistic Logic Networks (PLNs), Meta-Optimizing Semantic Evolutionary Search (MOSES), LLMs, or other algorithms of similar strength. Such synthesis allows for informational gaps to be filled, and for pieces of relevant information to be grouped together.

We therefore propose Sherlock, a system that will help bridge this gap. In this project, we will explore various methods for Attention Allocation and how they may apply to this problem of identifying the best algorithms for a particular selection of atoms. Our project contains two main parts:

  • Define a method for evaluating Attention Allocation systems that can supplement the need to identify which algorithms would be the most useful to apply to which atoms
  • Deliver a research paper that describes our novel architecture in detail and the benefits it affords to AGI

Team

Nate Lane, Computer Scientist and Language Model Expert - Project Leader

Nathaniel Lane graduated summa cum laude from Colorado School of Mines with a degree in Computer Science. He went on to get a Master’s in Computer Science from Montana State University by researching how neural networks can be used to predict whether a given peptide has anti-cancer properties. Since then, he has worked at MLabs on a variety of projects, including “Memory-augmented LLMs: Retrieving Information using a Gated Encoder LLM (RIGEL)” from RFP round 3. Nate will lead the project.

Nigel Farrelly, MLabs Data Scientist - Project Data Scientist

Nigel is an accomplished Data Scientist and Software Developer, with a proven track record in engineering solutions against complex requirements. Nigel holds a B.A. in Computer Science from Trinity College Dublin, Ireland, and combines his technical proficiency with a passion for problem solving, making him a valuable asset to the project. Nigel will assist Nate in developing a framework to support inference attention.

Challenges

This is an exploratory project, aiming at developing and demonstrating a novel framework for evaluating Attention Allocation, and we hope to illustrate a broad applicability to a wide variety of potential deployments. As such it is an ambitious remit for such a short project with limited budget.

We believe that the inherent risk is at least somewhat mitigated by the extensive experience of the project team members, who can also benefit from regular troubleshooting workshops with other members of the MLabs AI team. Furthermore, we invite discussion and collaboration from the wider AGI community, and trust that this will further reduce technical risk.

 

Open Source Licensing

MIT - Massachusetts Institute of Technology License

Links and references

Website

https://www.mlabs.city/

Proposal Video

Not Avaliable Yet

Check back later during the Feedback & Selection period for the RFP that is proposal is applied to.

  • Total Milestones

    3

  • Total Budget

    $30,000 USD

  • Last Updated

    5 Dec 2024

Milestone 1 - Correctness of Atom Attention Evaluation

Description

It is perhaps most vital that atoms of knowledge be correctly identified as “important.” As such for the first milestone of this project we will craft techniques to evaluate whether or not the proposed Attention Allocation system has done so correctly. This “correctness” determination will inform a “scoring” system wherein points are awarded when atoms are correctly given the level of attention they need while points are deducted when atoms are given an incorrect level of attention. This system will allow for custom criteria. For instance in some systems a false positive may be more burdensome than a false negative. Additionally some systems may care only if attention is given or not but others may need to consider levels of attention (such as short term or long term as in ECAN).

Deliverables

A white paper section describing how we judge and score correctness of atom attention allocation

Budget

$10,000 USD

Success Criterion

1. The scoring system gives a meaningful judgement on how well the atoms were selected for attention 2. The scoring system can be adjusted to meet the needs of the particular use case

Milestone 2 - Correctness of Algorithm Selection

Description

Next we must determine how well the proposed Attention Allocation system distributes tasks among the various algorithms it has access to. We will start by analyzing scenarios where it is known what the desired task distribution is. Once we have done so we will be able to create an objective scoring system that evaluates the efficacy of the task distribution. As before this system must also be adaptable to the given system’s needs and requirements.

Deliverables

A white paper section describing how we judge and score correctness of atom attention allocation.

Budget

$15,000 USD

Success Criterion

1. The scoring system gives a meaningful judgement on how correct the attention allocation system was in determining which algorithm fit the given data 2. The scoring system can be adjusted to meet the needs of the particular use case

Milestone 3 - Completed Paper and Conclusions

Description

At this point the most important parts of the paper and framework will be completed. We will wrap up the project by combining our findings into one cohesive paper wherein we describe our conclusions and what additional work may be done.

Deliverables

A final paper describing Sherlock - the attention allocation framework.

Budget

$5,000 USD

Success Criterion

We present our research and synthesize meaningful conclusions that help lay the groundwork for the development of attention allocation mechanisms

Join the Discussion (0)

Expert Ratings

Reviews & Ratings

Group Expert Rating (Final)

Overall

2.6

  • Feasibility 3.0
  • Desirabilty 2.8
  • Usefulness 2.3
  • Expert Review 1

    Overall

    1.0

    • Compliance with RFP requirements 1.0
    • Solution details and team expertise 1.0
    • Value for money 1.0
    Lack of expertise and naive proposal!

    Strong reject! The team poses very little expertise in the field of Attention allocation and control. While RFP asks to focus on evaluation framework development, proposal speaks on how to improve ECANs without even understanding what ECANs are and how they are used. Besides Nate is already submitted another weak proposal for PLN RFP.

  • Expert Review 2

    Overall

    1.0

    • Compliance with RFP requirements 1.0
    • Solution details and team expertise 1.0
    • Value for money 1.0
    Method to decide "which algorithm for which atom"

    "Define a method for evaluating Attention Allocation systems that can supplement the need to identify which algorithms would be the most useful to apply to which atoms". The contribution to Attention Allocation is not clear in this case, and the technique they plan to develop for this is not described. Hence only one star is given.

  • Expert Review 3

    Overall

    2.0

    • Compliance with RFP requirements 3.0
    • Solution details and team expertise 2.0
    • Value for money 2.0
    the approach seems quite naive, though professionally stated

    The approach described has the following properties 1) everything it describes implementing is there in ECAN already, though described in a different vernacular, 2) it seems oblivious to the difficulty of the assignment of credit problem (you can't just 'assign scores' to things that are important in a certain context ... the problem is how do you estimate the score to be given to something that's indirectly useful in a certain context... and then how do you evaluate different ways of doing this estimation...)

  • Expert Review 4

    Overall

    3.0

    • Compliance with RFP requirements 4.0
    • Solution details and team expertise 4.0
    • Value for money 3.0
    Proposal about focusing on attention allocation of processes, rather than just data.

    I may not completely understand the proposal, it seems to say that attention allocation is not just about data but also processes, which I agree. I suppose one can apply the same principles of attention allocation to data and processes, though they are going to be differences in the type of knowledge that is going to be inferred and used in both cases. I don't know if this proposal falls completely under the scope of that RFP, which is primarily about evaluation attention allocation and I find it also somewhat under-specified.

  • Expert Review 5

    Overall

    4.0

    • Compliance with RFP requirements 4.0
    • Solution details and team expertise 4.0
    • Value for money 2.0
    Credible team but little detail given on solution

    Vague proposal to evaluate. Need more information.

  • Expert Review 6

    Overall

    5.0

    • Compliance with RFP requirements 5.0
    • Solution details and team expertise 5.0
    • Value for money 5.0

    Solid proposal for an evaluation framework.

feedback_icon