Schools Bill 2022

On the 11th May 2022 the Government published its Schools Bill, providing the legislation required to implement several of the ideas explained in their earlier White Paper which I analysed in a previous blog post. On the 12th May the Bill was followed up with a raft of explanatory policy statements. I have been going through all these documents, and summarise here points of key importance or that stood out for me.

It is worth noting at the outset that the policy statements go well beyond what’s in the Bill, as they set out plans for intended standards to be set in the future, with the Bill simply enforcing these standards as they arise. As such, the Bill provides for quite sweeping powers for the Secretary of State to amend these standards and duties over time without the requirement for additional legislation.

1. Academies

1.1. Intervention in Existing Academies

Part 1 of the Bill deals with changes to the landscape of Academies in England. I would characterise the scope here as two-fold: firstly, the aim is to move many existing powers from contractual funding agreements into primary legislation, and secondly to be able to apply these powers at the Trust level (‘proprietor’ in the Bill) rather than only at individual Academy level.

Currently, the DfE’s powers of intervention depend on the version of the Funding Agreement and Articles of Association used by a particular academy, which depends on when they converted to Academy status – a patchwork which is both inequitable and very hard to manage. The Bill would place all Academies under the same legislative framework, which is to be welcomed.

Notices to Improve, an existing mechanism (formerly known as ‘Financial Notices to Improve’) seem to be envisaged as the main approach for ensuring change at a Trust or Academy without using the ‘nuclear’ Termination option. It’s proposed that these notices can be issued under wider circumstances than currently, e.g. “[if the SoS is satisfied] that there are significant weaknesses in the proprietor’s governance procedures” Part 1 6(1)(b). I think on balance this is a good idea: it is much better for the DfE to be able to use appropriate levers to effect change, especially in cases of inadequate governance, rather than have to fall back on Termination as the only route.

Along the same lines, Part 1 Section 7 includes powers to appoint or require appointment of directors to an Academy either due to failure to act on a Notice to Improve, a serious breakdown in governance procedures, or for safeguarding reasons. This happens by directing the Trust to appoint a named person or, if the DfE prefers, a person with named skills and experience. The accompanying briefing paper discusses using this power to create Interim Trust Boards, not unlike existing Interim Executive Boards for maintained schools causing concern. It’s interesting that similar provisions existed at academy level in early versions of the model Articles of Association of academies, see e.g. Article 62 allowing the appointment of ‘Additional Governors’ for very similar reasons, and the consequent Article 64 requiring existing governors to stand down in these circumstances. These no longer exist in the current model Articles, and I have always thought this odd. I am pleased to see these intervention powers returning, and perhaps more importantly all schools placed on an equal footing in this regard.

1.2. Academisation

The main new power over LA maintained schools is the power for LAs to apply for an Academy Order via a new Section 3A of the Academies Act 2010. This appears as Part 1 29(2) of the Bill. This seems to be the way the Government is implementing its stated aim to allow LAs to establish MATs (see my analysis of the White Paper for more detail). In areas where the LA is aligned with Government plans, we can expect to see beginning of the end of maintained schools over the 2023/24 school year, I expect. However, there is no obligation for LAs to apply for academy orders, so I believe we will still see a mixed economy of MATs, maintained schools, and isolated SATs well into the future, but perhaps with more regional variation than currently.

1.3. Grammar Schools

Grammar schools and schools with religious character get significant emphasis in the Bill. At least for grammar schools, I read this as primarily a desire to reassure them that the only way they will lose their ability to select is through a parental ballot, not through one of the sweeping powers introduced in the rest of the Bill: essentially the Secretary of State appears to want to explicitly limit his own powers in this regard without bringing in further primary legislation. This makes sense in the context of the White Paper, as a way to bring more grammar schools into MATs.

1.4. Not in the Bill

There are a few things I noted previously as mentioned in the White Paper that the Bill gives no real insight into, such as:

the requirement to (re)broker schools into strong MATs after two consecutive Ofsted reports below Good
the setting up of an “arms length” curriculum body
mechanisms for local governance (though this is mentioned in the explanatory briefing)
MAT inspection
collaboration standards for MATs

There is also talk in the accompanying briefings of improvements to complaint handling by academies, another issue I have commented on before and where I agree that significant improvements need to be made.

I suspect the broad powers introduced by the Bill are likely to be used as an “Enabling Act” to introduce these various points via the school standards provisions outlined in Part 1 1(2) of the Bill. We are told that the Academy Trust Standards will come to parliament separately (see explanatory briefing), and I look forward to reading these.

2. The Rest of the Bill

The rest of the Bill deals with school funding, school attendance, independent educational institutions, and various miscellaneous items.

I was involved in determining the school funding formula for Essex LA for many years, and I don’t read this portion of the Bill as particularly radical. Long-term readers of my blog will know that I’m not very happy with the approach taken to the National Funding Formula (see, for example, my 2017 consultation response and the links to earlier posts going back some years therein). But the new content in this Bill is essentially simply about making the default funding position the “direct” (previously known as “hard”) National Funding Formula, with the Secretary of State authorising exceptions (presumably because they recognise that actually one size really doesn’t fit all!). I’m pleased to see a continued role for Schools Forums in the proposed legislation, including around de-delegation for maintained schools, in Part 2 (3)(a).

The school attendance provisions introduce a duty on LAs to maintain a register of children not in school and to exchange information between LAs when children on the register move, as well as various provisions to help clamp down on illegal schools.

Conclusion

From my perspective, the meat of the Bill is Part 1 on Academies, standards, and interventions. I am cautiously optimistic about the measures proposed to intervene in existing academies. In particular, I welcome the move away from high stakes “hands off” versus “terminate” interventions. Whether a particular intervention is best made at Trust or Academy level is worthy of considerable debate, and I think practice is likely to evolve here over time, but it appears to me that if the Government really wants MATs to take active responsibility for their Academies, as discussed in the White Paper, the balance will move toward Trust-level intervention and towards softer forms of intervention.

I am less comfortable with the proposed amendments to the Academies Act 2010 to allow LAs to issue Academy Orders. There appears to be no mechanism to provide for the LAs to decide which MATs academised schools will end up in following one of these orders. While there is no requirement for LAs to issue Academy Orders to all their maintained schools, I do wonder what will happen in those areas which keep significant maintained schools: how will the Government achieve their stated aim of full academisation here? Future legislation, or increasingly constraining LAs ability to adequately maintain their schools via funding or other levers?

Finally, in order to make the intervention measures work well in practice, good resourcing of Regional Director (currently Regional Schools Commissioner) offices will be absolutely essential. It therefore somewhat unfortunate that these papers arrived in the same week as a drive to cut civil service jobs.

Keeping the Pipelines Full

On the 16th May, my PhD student Jianyi Cheng (jointly advised with John Wickerson) will present his most recent paper “Dynamic C-Slow Pipelining for HLS” at FCCM 2022 in New York, the FPGA community’s first in-person conference since the pandemic hit.

Readers of this blog may remember that Jianyi has been working on high-level synthesis techniques that combine the best of dynamic scheduling with the best of static scheduling [FPGA 2020,FCCM 2021]. The general principle underlying his work is to make the most of what information we have at compile time to develop highly efficient custom architectures, while leaving what we don’t know at compile time to influence execution at run-time.

A very common design pattern in hardware acceleration is the idea of C-slow pipelining. Pipelining tends to be taught early in undergraduate programmes, but C-slow pipelining rarely gets mentioned. The idea arises in circuits with feedback loops. The basic approach to pipelining doesn’t really work in this setting: although we can throw multiple registers into the circuit, potentially improving clock frequency at the cost of latency, just like with feed-forward circuits, we can’t then overlap computation to achieve improved throughput, unlike the feed-forward case, because of the data dependency corresponding to the feedback loop.

C-slow pipelining essentially says “OK, but you can use the spare capacity induced by the pipeline registers to overlap computation of independent streams of data, if you happen to have them available.”

Our new paper introduces a dynamic HLS flow for C-slow pipelining. This is particularly valuable in the context of a globally dynamic environment but where certain components exhibit static control flow and can be efficiently pipelined, for example some deep but predictable computation that must be repeated many times but with the arrival times and sources for this computation may change dynamically at runtime, a perfect fit for our prior work.

Jianyi presents a way to leverage the Boogie language and tool flow from Microsoft Research to automatically prove sufficient conditions for C-slowing to be correct. He is then able to introduce a new hardware component within the Dynamatic HLS tool that allows the schedule to “run ahead” to implement certain bounded out-of-order executions corresponding to C-slowing at the circuit level.

At the cost of a small area overhead in the region of 10%, this combined software analysis and hardware transformation is able to reduce wall-clock execution time by more than half compared to the vanilla dynamic scheduling approach.

If you’ll be in NYC in mid-May, go along and hear Jianyi’s talk!

Nonlinearity is Your Friend

My former PhD student Erwei Wang and I recently teamed up with some collaborators at UCL: Dovydas Joksas, Nikolaos Barmpatsalos, Wing Ng, Tony Kenyon and Adnan Mehonic and our paper has just been published by Advanced Science (open access).

Our goal was to start to answer the question of how specific circuit and device features can be accounted for in the training of neural networks built from analogue memristive components. This is a step outside my comfort zone of digital computation, but naturally fits with the broader picture I’ve been pursuing under the auspices of the Center for Spatial Computational Learning on bringing circuit-level features into the neural network design process.

One of the really interesting aspects of deep neural networks is that the basic functional building blocks can be quite diverse and still result in excellent classification accuracy, both in theory and in practice. Typically these building blocks include linear operations and a type of nonlinear function known as an activation function; the latter being essential to the expressive power of ‘depth’ in deep neural networks. This linear / nonlinear split is something Erwei and I, together with our coauthors James Davis and Peter Cheung, challenged for FPGA-based design, where we showed that the nonlinear expressive power of Boolean lookup tables provides considerable advantages. Could we apply the a similar kind of reasoning to analogue computation with memristors?

Memristive computation for the linear part of the computation performed in neural networks has been proposed for some time. Computation essentially comes naturally, using Ohm’s law to perform scalar multiplication and Kirchhoff’s Current Law to perform addition, resulting in potentially energy-efficient analogue dot product computation in a physical structure known as a ‘crossbar array’. To get really high energy efficiency, though, devices should have high resistance. But high resistance brings nonlinearity in practice. So do we back away from high resistance devices so we can be more like our mathematical abstractions used in our training algorithms? We argue not. Instead, we argue that we should make our mathematical abstractions more like our devices! After all, we need nonlinearity in deep neural networks. Why not embrace the nonlinearity we have, rather than compromise energy efficiency to minimise it in linear components, only to reintroduce it later in activation functions?

MNIST classification error trading off against power consumption

I think our first experiments in this area are a great success. We have been able to not only capture a variety of behaviours traditionally considered ‘non-ideal’ and harness them for computation, but also show very significant energy efficiency savings as a result. You can see an example of this in the figure above (refer to the paper for more detail). In high power consumption regimes, you can see little impact of our alternative training flow (green & blue) compared to the standard approach (orange) but when you try to reduce power consumption, a very significant gap opens up between the two precisely because our approach is aware of the impact this has on devices, and the training process learns to adapt the network accordingly.

We’ve only scratched the surface of what’s possible – I’m looking forward to lots more to come! I’m also very pleased that Dovydas has open-sourced our training code and provided a script to reproduce the results in the paper: please do experiment with it.