A²ISR WP2 - Distributed AI (PED) at the Edge

Tender description

About adding context and requirements test Pre-market engagement N/A Work done so far The DAIS ITA has delivered world leading inter-disciplinary scientific outputs, enabling generation after next information integration and exploitation, for distributed coalition multi-domain operations in dynamic and contested environments. It has developed the science underpinning (a) Resilient Coalition Networks for degraded tactical networks and (b) Artificial Intelligence (AI) in Contested Multi-Domain Coalition Operations. This work is available at https://dais-legacy.org/. Further research and development has been conducted for MOD under the SERAPIS framework (a) workflow composition and orchestration in task AII107 and (b) federated learning in task AII115. The output of these two tasks will be available as GFX after contract placement Which phase the project is in Alpha Existing team The supplier will be working with a mixed team of Dstl specialists, stakeholders from across defence and other contracted parties under a wider project. Principal engagement on this task will be with the: Concept lead for AI anaytics at the edge, and their techncial team (incuding suppliers from industry) Project Technical Authority Lead Technical Reviewer Military Advisor External stakeholders in Defence and other Dstl projects Address where the work will be done Largely remote but with regular and frequent attendance at Dstl Porton Down (SP4 0JQ) and regular travel to stakeholders around the UK. It is expected the team could be nationally based, but the client location is predominently southern based. Working arrangements It is expected that the supplier and Dstl team will meet virtually or face to face at least once a month to review progress; with other major events programmed in as required. Regular interaction, face to face and virtual will also be required with other activities in the wider project. Provide more information about your security requirements: Security Check (SC) Provide more information about your security requirements (optional): Security Check level of clearance is required for majority of team. The supplier is expected to process and hold their own clearances. Clearances can be managed by 3rd parties. The authority will not hold or sponsor clearances. Latest start date 2023-08-14 Enter the expected contract length: 1 year, 7 months and 17 days Special terms and conditions A combination of DEFCON 703 and DEFCON705 shall apply throughout. A full Cyber Risk Assessment (DCPP) will be required prior to contract. Write the term or acronym: AI Write the term or acronym: PED Write the term or acronym: ISR Write the term or acronym: TRL Write the term or acronym: DDIL Write the term or acronym: VR Write the term or acronym: CCR Write the term or acronym: GAN Explain the term or acronym: Artificial Intelligence Explain the term or acronym: Processing Exploitation and Dissemination Explain the term or acronym: Intelligence Surveillance and Reconnaisance Explain the term or acronym: Technology Readiness Level Explain the term or acronym: Denied, Degraded, Interrupted, Low bandwidth Explain the term or acronym: Vertical (major) requirements Explain the term or acronym: Cross Cutting requirements Explain the term or acronym: Generation After Next Are you prepared to show your budget details?: Yes Indicative maximum: £1.575M (ex VAT) Provide further information: Total limit of liability is £1.575M (ex VAT). We would like costed Option 1: We would like costed options to extend the duration by 25% and 50%, currently there is no guaranteed funding for this. As per technical question 5 - We would like costed Option 2: A plan for delivery and management, if additional funds become available for FY24/25, for continued research exploitation, this should be in region of £0.75M. Confirm if you require a contracted out service or supply of resource Contracted out service: the off-payroll rules do not apply Summary of work Cutting edge research on distributed edge AI and analytics in tactical environments is required to support future ISR capabilities; including (1) a prototype Edge (inc. Federated and Frugal) AI production facility (TRL 6) ; (2) self-healing distributed analytic workflows (TRL 4); (3) international experimentation. This is a 2 year task. Where the supplied staff will work No specific location (for example they can work remotely) Why the work is being done Developing the Science & Technology required to enable distributed analytics and Artificial Intelligence (AI) on edge computing platforms (from servers to embeded devices) in adversarial and degraded communications environments and with partners (including international). It contributes to addressing the core capability challenge of “pervasive, full spectrum, multi domain Intelligence, Surveillance and Reconnaissance (ISR)” identified in the 2020 MOD Science and Technology Strategy, and seeks to work with cutting edge and innovative suppliers in industry and academia. It is focused on maturing emerging concepts (TRL 2-4+); and must be completed within 2 years. The business problem you need to solve Development of the Science and Technology covering 4 inter-related needs and 5 requirements: (1) prototype (TRL 6) an evolving Edge AI (including Federated and Frugal AI) production and test facility (producing TRL 4 outputs) – referred to as requirement VR1 below; (2) develop and test AI driven services enabling autonomous and self-healing composition and orchestration of distributed analytic workflows in a deployed tactical environment (TRL 4) – referred to as requirement VR2 below; (3) support integration and experimentation with partners, including international partners – referred to as requirements CCR2 & CCR3 below; (4) bring the envisaged future to life for stakeholders – referred to as requirement CCR1 below. Developing this S&T in order to support the tasks required of a future ISR enterprise (e.g. processing, exploitation and dissemination of ISR data) operating in a distributed manner. First user type: Defence Scientist Enter more details about this user type: As a Defence scientist I need to work with the cutting edge research and development talent to deliver understood, feasible and affordable solutions to the challenge of enabling distributed analytics and Artificial Intelligence (AI), so providing distributed processing, exploitation and dissemination of information, on edge computing platforms (from servers to embeded devices) in adversarial and degraded communications environments and with partners (including international), so that Defence can equip future UK ISR and Intelligence capability with battle winning technology that supports an enduring state of Information Advantage. Questions and Clarifications 1. We believe it would support the submission of more cost effective Tenders from all of the Tenderers if DSTL could clarify which of the 26 DOS 6 Call Off Schedules and which of the 11 DOS 6 Joint Schedules will be applied to the A2ISR. Additionally, we consider it would be of further benefit to the ITT if the intended Terms, which are to be reflected in the DOS 6 required Order Form could also be clarified. Draft schedules will be made available following the evaluation of stage 1 Last Updated: 25 May 2023, 14:35 2. Can Dstl advise whether the current industry team involved in the research is expected to respond to the ITT? VR1 relates to, but is very different to, a strand of work undertaken through SERAPIS Lot 4 task AII115 by a supplier. VR2 relates to further development of two strands of work undertake through SERAPIS Lot 4 task AII107 by a team of suppliers. The outputs of these, including code, will be available as GFX. Dstl has no expectation that the suppliers of these earlier pieces of work will, or will not, respond to this ITT. The suppliers of previous work have not been involved in writing these requirements. Any members of the A²ISR project office are required to declare conflict of interest and how they will manage that conflict, if responding to other requirements. The authority will then review these plans and respond regarding whether they can be part of the bid team. Last Updated: 25 May 2023, 14:27 3. With a given model developed, data availability will vary which, under the scope, will require an iterative/ recursive approach to training the model/s e.g. a need has been identified through the Wizard-of-Oz or UCD approach, but there is no/ limited data; 1. What is the plan to collect the data/ train the model? Pool-based/ streaming? (I imagine there will be a variety of approaches - as mentioned incorporating public sourced data in CCR1 - but how does this tie in with the requirements of the edge device? Are we free to choose, given the situation, the best approach or are there some scenarios that will be strict on their required processes?) 2. Are we providing the tools to the user, to decide on a given scenario what is the best approach? (Through the SDK's?) 3. Will we be incorporating data science lifecycle components, building model monitoring systems that detect model degradation? Or will the models be periodically / stream-based updating? 40-1 Assuming this refers to VR1 see answer above to Q7 and Q23. If this refers to CCR1 and the outputs of Wizard-of-Oz and UCD approaches then there is more freedom to envisage future processes. CCR1 use cases are expected to be realistic / representative; but may use open source proxies (e.g. Transport for London data). Likewise edge devices may be off the shelf – illustrative rather than GAN devices. 40-2 Yes the scope should be to inform and assist the user. If referring to VR1 (inferred from reference to SDK) then the production facility will need to include processes and tools for selecting best approach. If referring to VR2 then expect a user-on-loop control approach. 40-3 At this stage, it is expected the models with be periodically updated. Monitoring of model performance is expected to be a feature of a fully developed system: the need to implement such monitoring systems to meet the requirement of the task should be recorded as part of the evidence based decision making on the task. Last Updated: 25 May 2023, 14:25 4. Are Dstl content with supplier teams that comprise of more than one organisation. Dstl are keen to have proposals from multi-organisation teams, as emphasised by the ‘cultural fit’ questions. Last Updated: 25 May 2023, 14:24 5. Dstl have indicated that there is a SAL for the work, can that please be provided. Yes – this will be provided following the evaluation of stage 1. Last Updated: 25 May 2023, 14:23 6. Can Dstl state whether there are any restrictions on non-UK nationals being involved in the work. If so, are 5-Eyes nationals acceptable. SC clearance is required for the majority of the team. From a security perspective, if the work stays at official and has no nationality caveats then a non-UK national can work on it. If the work raises in classification then all staff will need SC clearance. Non-UK nationals will not be able to attend ‘in person’ meetings at Porton Down. Research worker forms will be needed for non-UK nationals and if they are rejected an alternative resource will be needed. Generally EU nationals are accepted. Last Updated: 25 May 2023, 14:22 7. Can Dstl confirm that export permissions and licences will not be required for the work. The supplier will not be expected to arrange export permission or licences. Permissions and licenses, if required, will be facilitated under existing defence agreements. Last Updated: 25 May 2023, 14:21 8. Can suppliers assume that Dstl will manage and arrange any MOU or similar agreement for international collaboration and working. An MOU, and project arrangements (PA) are already in place for international working. Last Updated: 25 May 2023, 14:18 9. Please clarify whether international travel and expenses should be allowed for in the proposal. No international travel is expected. UK wide travel is expected. Last Updated: 25 May 2023, 14:17 10. What is the definition of acceptance from MoD for the contract activities / deliverables? Acceptance is as described in the SOR document that has been emailed out. Last Updated: 25 May 2023, 14:17 11. For requirement VR2.3 you state "integrate with related Comms & Networks and Cyber Defence programme projects that provide Service Discovery, Communications Management, Translation, Service Management and IDAM (Identity & Assess Management) middleware services". Please can name them / provide a number of these systems expected to integrate too? There is parallel research into next generation in range of other projects into these services: many of these are using the OCTANE environment. As noted above: the ethos needs to be of collaboration and openness to working with other suppliers working under contract to Dstl. The most important other middleware services are those listed: namely Service Discovery, Communications Management, Translation, Service Management and IDAM (Identity & Assess Management). Last Updated: 25 May 2023, 14:16 12. What MoD system are the service request taken from that them are used to create the workflows? There is parallel research into next generation systems for generating service requests. The service requests are not taken from a current MoD system. Supplier is expected to develop / use appropriate workflows per CCR1 bringing it to life to demonstrate representative workflows. Open source / proxy data/ concepts may be used to avoid confidentiality issues. Last Updated: 25 May 2023, 14:15 13. Will DART registration and other MOD compliance criteria need to be meet for delivery of a solution? Or will MoD cover this activity? Suppliers are not expected to apply DART to their own hardware. If the authority need to use the hardware at a later date on this contract, the authority will complete the DART assessment. Last Updated: 25 May 2023, 14:15 14. Do approved integration approach /API's exist to all the external MoD or third party systems / services / micro services that will need to be access for this project? No. There is, as noted in previous responses, considerable parallel research and development. Integration only needs to be sufficient to meet the requirements of the task (e.g. some services can be stubbed). Last Updated: 25 May 2023, 14:14 15. Are there agreed working arrangement between MoD project teams on how to work together for this project or does that have to be created? All people working on the project will be under contract (or part of the MOD community), we can therefore share the work as GFX (depending on IP conditions) with suppliers under contract. The whole ethos needs to be collaboration and openness to enable the research and design of a generation after next Intelligence, Surveillance, Reconnaissance Enterprise, as emphasised by the ‘cultural fit’ questions. Last Updated: 25 May 2023, 14:13 16. Are there existing formal commercial relationships agree with all the Academia groups with MoD and that may be engaged on this project, and what does this cover; how will this work with the supplier - will it be a direct relationship to them or through MoD? For related academic/supplier work, there will be a separate contract with the relevant institute/company. Within all of our contracts we are asking for suppliers to work collaboratively in this area of research. Last Updated: 25 May 2023, 14:12 17. What edge devices to integrate with - custom or off the shelf? Suitable to demonstrate the specified outcomes at the specified TRLs: thus can be custom or off-the-shelf as required. Some off-the-shelf low SWaP devices may be available as GFX. Last Updated: 25 May 2023, 14:11 18. What volumes of data will be required and from what sources? Will this need to be stored in OCTANE? Sufficient to achieve the specified outcomes at the specified TRLs. While the work is ongoing the data sets would be on the supplier hosted environment, ideally with access by the authority. At the end of the contract it would need to be transferred to the authority. As development progresses and, ideally but other solutions will be considered, integration and deployment on OCTANE occurs then data would need to be used on OCTANE, but may be transient for the duration of an experiment / demo if long-term storage is elsewhere. Last Updated: 25 May 2023, 14:10 19. How many data sources will need to be integrated and from what type of systems - are there integration already in existence that can be re-used? Sufficient to achieve the specified outcomes at the specified TRLs. See also response to question surrounding federated models. Last Updated: 25 May 2023, 14:09 20. What processes are you expecting the toolsets and frameworks to deliver - e.g. dev packages, code repo, testing tools, requirements management, configuration packaging? Assuming this is referring to VR1; then a suitable (evidence based) set to meet the requirement. Last Updated: 25 May 2023, 14:07 21. What is the sign of process for use cases / user stories Sign-off will be with Dstl Technical Partner. Last Updated: 25 May 2023, 14:06 22. What tool sets have been used to create the AI algorithms for the existing research undertaken A range of toolsets have been used by different projects; frequently the toolsets are used by a supplier to achieve the specified outcome and are not owned by the authority. Federated Learning has been demonstrated using various open source CNN and SGD libraries Workflow orchestration has made use of serverless workflow pattern and hypergraph auction algorithms. Further details can be made available as GFX once under contract Last Updated: 25 May 2023, 14:06 23. Who is the MoD audience from which user stories / use cases will be generated from? is this already in existence? We are looking at the generation after next for the intelligence surveillance and reconnaissance enterprise, the user is the intelligence, surveillance and reconnaissance community. Last Updated: 25 May 2023, 14:05 24. What is the character count for the technical, cultural and social value questions? There is no character count in place for the stage 2 technical, cultural and social value questions. We would expect your responses to be incorporated into your technical proposal. Last Updated: 25 May 2023, 14:04 25. What is the 'OCTANE' environment and what hardware/sofftware does it include? OCTANE (Open CIS (communications information systems) Test and Experimentation) is a third party hosted sandbox environment that is run for Dstl, allowing external parties to connect remotely and perform online experimentation and collaboration. It is a virtualised environment supporting containers and relevant tools as CORE/EMANE (Common Open Research Emulator / Extendable Mobile Ad-hoc Network Emulator) suitable for representing DDIL environments. Last Updated: 25 May 2023, 14:00 26. Would the code repository be on customer managed environment, or on a hosted environment? While the work is ongoing the code base would be on the supplier hosted environment, ideally with access by the authority. At the end of the contract it would need to be transferred to the authority Last Updated: 25 May 2023, 13:59 27. Will there be a requirement for Dev Sec Ops within the project? It is likely that meeting VR1 will require demonstrating Dev Sec Ops; but otherwise it is not required within the project. Last Updated: 25 May 2023, 13:59 28. Is there a required style guide towards any and all documentation that is to be supplied alongside the software? In general, the following guidance should be followed in preparing code, APIs and associated documentation: https://www.ncsc.gov.uk/collection/developers-collection/principles/produce-clean-maintainable-code https://www.gov.uk/guidance/gds-api-technical-and-data-standards and https://www.gov.uk/guidance/how-to-document-apis For reports - Defence research report specification Last Updated: 25 May 2023, 13:58 29. Can the Authority confirm if VR3 and VR4 handled by separate teams, or is it a consideration for Fujitsu once VR1 and VR2 are complete? Separate competitions are being/will be run for VR3 and VR4 (under the Serapis framework), it is unknown who will currently delivery these requirements. Last Updated: 25 May 2023, 13:57 30. Can the Authority confirm if analytics should also include the central hub of the federated system along with the edge analytics? Yes: assuming the question is asking if analytics on the performance/behaviour of the federated learning systems as a whole is important. Last Updated: 25 May 2023, 13:56 31. Can the Authority confirm if there is an expectation to attempt various attacks to prove security of the federated models? This task is not seeking to prove the security of federated models against attacks; an awareness of risks is sufficient for the task. Last Updated: 25 May 2023, 13:55 32. Should analytics be set within DSTL’s dashboard, or ad-hoc? Current plan is for Ad Hoc Last Updated: 25 May 2023, 13:54 33. CCR3 - how will timelines be set up for the international collaboration? There is an international demonstration/experiment in October 2023, but we will be mainly observing, understanding the work and components being used by other organisation, and building relationships with the community. The next international demonstration is provisionally planned for October 2024 where we will want to be more integrated. Last Updated: 25 May 2023, 13:53 34. Can the Authority confirm what hardware limitations should we design around to best capture what would be used within the DDIL environment? As we are looking at generation after next, it is uncertain what these will be. We are looking for novel and innovative thinking for pushing the boundaries, and developing new systems. Last Updated: 25 May 2023, 13:52 35. Can the Authority confirm if we will know ahead of time what sort of data the federated models will have to be trained to handle? The current expectation is that the focus will be on imagery data, due to ease of organising experiments with international and other partners, but this is not fixed and is open to negotiation with suppliers. Suppliers should indicate their availability of or ability to generate/create/synthesise suitable datasets. Last Updated: 25 May 2023, 13:51 36. Please confirm that only Essential and Nice-to-have criteria are required by 02/06/23 Confirmed, please submit your responses to the Essential and Nice-to-have criteria by 2nd June. Last Updated: 25 May 2023, 13:47 37. Please can you confirm that 750 characters is per answer and not in total? Each response should not exceed 750 characters in length (the equivalent of about 150 words), this includes spaces. You will not be excluded if you go over this count, but we will only evaluate the first 750 characters of the response. Last Updated: 25 May 2023, 13:45 38. We can't open the SOR, it says unreadable content. Can you check and re issue The SOR has been reissued to all Lot 1 Suppliers by email, along with the bid pack. Last Updated: 25 May 2023, 12:07