>

How preprint servers gave us an edge in an invalidation search

As a searcher, each of us has an array of tactics and strategies at our disposal. Implementing the right tactic at the right resource could open doors to the prior art which you thought might not have existed despite a thorough search. 

While each of the searchers has their own set of strategies they like to employ for a given situation, in my opinion, looking up at the right resources can give you an edge. 

Take preprint servers, for instance.  Did you know pre-print servers can be a fantastic place to find killer prior art

I had no idea either until we started working on our client Sam’s project. I will tell you all about it today but first, let’s address the elephant in the room, shall we?

What are preprint servers and how can they help in invalidation searches?

Preprint servers are online repositories that serve as a platform for researchers/scholars to instantaneously upload their work. Unlike additional or renowned academic journals, these papers are not peer-reviewed before being published. They only undergo basic screening checks for plagiarism.

Not to give any spoilers early on, but one of these preprint servers gave us the kind of info in the case which was not accessible anywhere else. 

What did we find and how did that info help? Read on to find out. 

Sam wanted us to invalidate a Neural Networks related patent in a very tight deadline

Sam, a senior IP counsel, at a Fortune 500 corporation had recently reached out to us with a very interesting situation. He wanted us to conduct an invalidation search project albeit at a very tight deadline. 

Sam had consulted a few IP firms for this project already, but none of them were able to find any solid art to make this case. So, he was on a tight deadline. Sam had to submit the results in a week, which meant we had only 5 days at max to share the results with him. This case was big for his corporation and he had heard about us so he wanted to give a final try. 

The technology of the subject patent which involved the nuances of optimizing deep neural networks didn’t make our job any easier either. This was a tough one. But we like challenges.  Our plan was to explore every possible avenue, albeit in a smart manner. 

Grit alone wasn’t enough here, this case needed a strategic approach. Given the short timeline, three researchers – we were all working in parallel – were assigned the task to find a killer prior-art to this case. We had a lot of factors working against us and we needed a quick head start. 

For that reason, we decided to consult the product development team at GreyB. 

It should also be noted that given the NDA we signed with Sam, and our clients in general, we take extreme caution to not disclose any confidential information while any such inter-team communication happens. Confidentiality above all, that’s the thumb rule with us. 

Well coming back to our PD team, they had built amazing AI and ML tools and could quickly help build an understanding of this tech. They shared the basics of neural networks technology with us. One of the team members, Kushal Brahmin, also shared that arxiv.org is a hot place to submit research work related to ‘neural networks’.

Back to the case, I had a lead for further exploration in case we don’t find anything through conventional methods. 

The search for prior art starts – But it’s no cakewalk

We had started our search employing conventional methods – Looking up all the relevant keywords on commercial patent databases, employing every individual tactic the three of us were aware of. But despite an exhaustive 3-day exploration, we had nothing. We could not find any document discussing the novelty. 

This situation has never been easy to live with, but we were not going to give up. We had two more days.  

Any of the references we found, and those shared by Sam from other vendors were all after the cut-off date. But that’s not the end of the world. 

In our experience, we have noticed that such leads if dug deeper could lead you to something significant.

As a next step, we backtracked all the leads we had found and those shared by Sam, which were after the cut-off date. Among those, there was one research paper disclosing novelty – “EIE: Efficient Inference Engine on Compressed Deep Neural Network”. 

However, unlike the other few leads, it wasn’t submitted by the inventors of the target patent. This was worth a shot. This is when I recalled our AI/ML expert’s suggestion to look into arxiv.org. What happened next was one of the best moments of my life as a researcher. 

How details on a preprint server helped us get our hands on killer prior art?

As per the details provided on IEEE Xplore, this paper was presented at a conference – 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA) that took place in June of 2016. And this paper was published in IEEE in August 2016. Both of these dates were after the cut-off date for this search:

The amazing part is what we found on arxiv.org

The paper was first available on February 4th on this server! Same paper, same content, but a gap of about 6 months in its publishing. This proved to be a game-changer in our search. 

We also found another document in the next couple of hours titled Reducing the Model Order of Deep Neural Networks Using Information Theory. For this paper, the research gate shows the publication date to be in July 2016, but on arXiv, it was May 2016. We found it by searching on arXiv.org and used it as a supplement needed for the case. 

Once we had all the information we needed, we drafted an email and shared our findings with Sam. There were still a few hours left before the deadline and Sam was elated with the quality of art we were able to find.  Sam shared this over email with our sales team: 

“Cost-saving is important and I often use other vendors for their quick and cheap searches. But results like this remind me why we use firms like GreyB. You guys were totally worth the extra cost. Keep up the good work.” – Sam. 

A Note on preprint servers

Well, to be honest, I feel our software team guys deserve a bit of the credit too. If they hadn’t shared anything about arXiv, we would have had to share a “No results found” report too. This case added one of the best search resources to my arsenal.  Now, I religiously explore preprint servers in most of my searches, and it works in our favor most of the time. 

I am not alone. In fact, a few examiners use this strategy too. A few days ago, I read the following comment from a WIPO examiner on Reddit, which I kinda would have missed had it not been for this case. 

WIPO Examiner on reddit

Yes, preprint servers are a great search resource and you should add them to your arsenal too. The good news is there are a good number of such servers available online. Some of the renowned ones in addition to arXiv include  BioRxiv, Chemrxiv, The Winnower, PsyArXiv, PrePubMed.

Concluding Remarks

Before I conclude, I have two cents, or to better put it, a handy tip you should know while conducting a search.

Whenever you are searching for prior art for a patent that comes from a university, or get a close research paper submitted after the cut-off date:

Chances are that the inventor(s)/author(s) might have described the work in a research paper first since they are more focused on conferences than on patents. Thus, exploring similar documents/articles in preprint servers would yield fruitful results in such cases. 

That’s my tip, y’all. Until next time!

Want to read more such cases where we found prior art in unconventional places? Head here next. 

Read quite a bit and want to engage us in a search project? Send us a quick message and we will get talking: 

Get in touch
Authored by: Himanshu and Ankush Sharma, Search Team. 

Leave a Comment