Video: Radenso Inside Look - Black Ops Engineering: Part 1

TechnoRave

PSL +5
Intermediate User
Joined
Jan 5, 2019
Messages
548
Reaction score
726
Location
#850Strong - Gulf Coast of Florida
This has been suggested many times already, great idea but I don't trust everyone (sorry, not sorry).
Well then make only a special Theia for your employees to do the recordings. Honestly I guess I’m missing the point of trust? Trust someone will record bogus information? Or tamper with it? Help me understand

If not. No biggie.


Sent from my iPhone using Tapatalk
 

NorEaster18

Probably stuck in traffic...
Advanced User
Joined
Sep 5, 2018
Messages
930
Reaction score
1,956
Location
New England
Beyond cool what is being shown in the video. And on top of that, the video editing in this is fantastic! Very enjoyable to watch, from the "conversation" style bits to the cut scenes. I am more and more excited for Theia everyday. Great job guys!
 

Jon at Radenso

Manufacturer
Advanced User
Premium Member
Manufacturer
Joined
Oct 9, 2012
Messages
1,704
Reaction score
12,245
Age
31
Beyond cool what is being shown in the video. And on top of that, the video editing in this is fantastic! Very enjoyable to watch, from the "conversation" style bits to the cut scenes. I am more and more excited for Theia everyday. Great job guys!
Video editing is so hard! And time consuming! It's just one more skill that I am forcing myself to try to learn - when you are a small company, you kind of need every employee to know a little bit about everything since we can't afford specialists. I am sure I made tons of mistakes and there is so much more I wish I could do better with, but until the $$$ comes in in from Theia and I can hire a real professional you guys are stuck with my edits! Thank you for the kind words, hopefully we continue to learn what works and what doesn't and always keep improving each week.
Post automatically merged:

Well then make only a special Theia for your employees to do the recordings. Honestly I guess I’m missing the point of trust? Trust someone will record bogus information? Or tamper with it? Help me understand

If not. No biggie.


Sent from my iPhone using Tapatalk
It's not even so much trust as the difficulty of getting a clean, unpolluted recoding. Believe it or not, since we are using really advanced DSP we can actually see signals below the noise floor. This is problematic if we let most people record and send us samples, because there is no easy way for the average user to see if a signal below the noise floor is present or not visually. But if it is, it will screw up the AI inferencing.

Basically, we need to be super, super, super careful about the quality of our training data or it all goes to hell.
 

Min Diesel

Manufacturer
Beginner User
Manufacturer
Joined
Nov 5, 2018
Messages
80
Reaction score
282
Location
My Desk
Video editing is so hard! And time consuming! It's just one more skill that I am forcing myself to try to learn - when you are a small company, you kind of need every employee to know a little bit about everything since we can't afford specialists. I am sure I made tons of mistakes and there is so much more I wish I could do better with, but until the $$$ comes in in from Theia and I can hire a real professional you guys are stuck with my edits! Thank you for the kind words, hopefully we continue to learn what works and what doesn't and always keep improving each week.
Post automatically merged:



It's not even so much trust as the difficulty of getting a clean, unpolluted recoding. Believe it or not, since we are using really advanced DSP we can actually see signals below the noise floor. This is problematic if we let most people record and send us samples, because there is no easy way for the average user to see if a signal below the noise floor is present or not visually. But if it is, it will screw up the AI inferencing.

Basically, we need to be super, super, super careful about the quality of our training data or it all goes to hell.
As Jon says... “garbage in, garbage out”. So, so, so true. Looking at signals in Phosphor, it’s glaringly obvious as to how ‘clean’ the data needs to be. However I NEVER would have known exactly how OCD to be about this until I stared at Phosphor for hours on end. A clean collection is the only collection.
 

sdrawkcaB

Defender of Light 📶
Advanced User
Lifetime Premium Member
Joined
Mar 6, 2018
Messages
5,008
Reaction score
9,772
Location
Here
Maybe there can be a “mark this spot” type of app, similar to AT&T’s app for reporting areas of poor coverage. It basically takes a few quick tests of telemetry data and then reports it to AT&T. The more user reports of an area, then the higher priority that location gets for the local network engineers to come out and assess the situation, thus the sooner they come and determine how to improve the area.

Radenso could offer us something similar. We realize that based on what you’ve said here, that us contributing raw signal data for training just isn’t feasible. That said, what we CAN do is simply collect basic telemetry data, and submit it to Radenso along with the location of the false positive (or false negative, be it the case). Then you guys can have the aggregate data and see what locations are being reported the most often, and now choose to dispatch the Radenso R.A.M., aka the Radio Analysis Mobile; your mobile testing and data collection equipped vehicle. Travel out with the RF engineers to the highest priority locations and sample the radio environment, identify and record good samples of the reported source, and take those back to train for false positives or negatives.

*Jon puts on mask and cape*
Jon: “Randy! To the Radenso RAM!”

429F7BD7-FF4D-4269-BE52-6CDF7AD94248.gif
 
Last edited:

TechnoRave

PSL +5
Intermediate User
Joined
Jan 5, 2019
Messages
548
Reaction score
726
Location
#850Strong - Gulf Coast of Florida
Video editing is so hard! And time consuming! It's just one more skill that I am forcing myself to try to learn - when you are a small company, you kind of need every employee to know a little bit about everything since we can't afford specialists. I am sure I made tons of mistakes and there is so much more I wish I could do better with, but until the $$$ comes in in from Theia and I can hire a real professional you guys are stuck with my edits! Thank you for the kind words, hopefully we continue to learn what works and what doesn't and always keep improving each week.
Post automatically merged:



It's not even so much trust as the difficulty of getting a clean, unpolluted recoding. Believe it or not, since we are using really advanced DSP we can actually see signals below the noise floor. This is problematic if we let most people record and send us samples, because there is no easy way for the average user to see if a signal below the noise floor is present or not visually. But if it is, it will screw up the AI inferencing.

Basically, we need to be super, super, super careful about the quality of our training data or it all goes to hell.
Thank you for taking the time to explain this. Makes sense. Keep up the awesome work!!


Sent from my iPhone using Tapatalk
 

V1Jake

Advanced User
Premium Member
Joined
Oct 31, 2010
Messages
5,509
Reaction score
5,792
Location
On the highway
Exactly plus I remember the beginning days of Trapster... it was false alert city. EVERYONE in Cincinnati was marking traffic cameras as red light cameras.
I remember that... had to delete soooooooooo many of them when moderating
 

poolmon

Advanced User
Lifetime Premium Member
Joined
Oct 26, 2013
Messages
726
Reaction score
1,127
Just trying to clarify my own understanding with respect to these new wrong-way emitters shown (nice video).
As I understand Theia, it does a combination of two things simultaneously via AI:
1. Mainly, it picks up only known and profiled radar emitters (white listed).
2. Also, it has the capability of specifically rejecting known false emitters (black listed), presumably to fine tune the white listed emitters and the probabilities associated with those platforms.

If this new emitter is not yet profiled, how can it be picked up as it will not be on the white list? I do understand how if there were a Legacy mode it would then be a useful filter for that mode. Thanks in advance for any clarification as we are all learning a lot as Theia progresses.
 

Bloovy One

110,000+ miles with JBV1!
Advanced User
Lifetime Premium Member
Joined
Feb 8, 2016
Messages
4,163
Reaction score
8,125
Location
Western NC
It might not be wise for any detector to treat an unknown signal as a blacklisted signal. Those should cause an alert if they are unknown. Theia will need as much information as possible to protect it's user without alerting to every unnecessary device out there.
 

Godowsky17

PSL +5
Intermediate User
Joined
Apr 24, 2019
Messages
388
Reaction score
849
Location
NY
It might not be wise for any detector to treat an unknown signal as a blacklisted signal. Those should cause an alert if they are unknown. Theia will need as much information as possible to protect it's user without alerting to every unnecessary device out there.
I’d imagine there will be different modes you can select - for example, one might alert only to known police radar sources, while another might alert to anything that hasn’t been specifically coded in as a false alert.
 

Bloovy One

110,000+ miles with JBV1!
Advanced User
Lifetime Premium Member
Joined
Feb 8, 2016
Messages
4,163
Reaction score
8,125
Location
Western NC
And that was mentioned by Jon early on. :) Still, ignoring an unknown signal is bad practice and can do severe damage to a detector's reputation. I suspect that such a mode will not exist in Theia.

The main reason some other detectors do behave that way is because there's a disconnect between what the company that sells them wants and what the manufacturer is providing. The manufacturer doesn't drive US roads in those scenarios and it's not their job to fix an issue ahead of time that wasn't part of the build order. :)
 

Heywood

Learning Something New, Still Dying Stupid
Advanced User
Lifetime Premium Member
Joined
May 12, 2016
Messages
4,250
Reaction score
8,172
Location
Alberta Canada
There are only so many guns approved for enforcement in North America.
There are only so many guns still in production.
There are so many gun that can still be serviced.

The ones approved for enforcement are the ones I want to be alerted to.

These “Wrong Way Signs” are more a curiosity. I think it’s a good idea to be able to set the detector to alert known approved law enforcement guns.

You can’t reasonably expect that your going to go out and record every single false alert signal known to mankind when you can focus on the short list of approved radar guns approved for enforcement.

Approval from the FCC doesn’t happen over night. Anything new will have to have approval and Radenso will probably know about it before it hits the market. It’s not like patents.
 

CarefulDriver

Learning to Fly
Beginner User
Joined
Jun 13, 2019
Messages
129
Reaction score
82
It's not the world end if it falses couple of times in year. I hope we will have option to make it alert to unknown signals. Heywood is indeed correct that it's practically impossible to record every signal ever known to mankind. Better safe than sorry.
 

poolmon

Advanced User
Lifetime Premium Member
Joined
Oct 26, 2013
Messages
726
Reaction score
1,127
I definitely agree that's its best to put in as many known emitters both black and white for maximum accuracy but I was just curious if it where not identified as a white listed emitter in the first place, why it would show up at all as a false? Just trying to understand the science a little better . .
 

Deacon

TXCTG
Advanced User
Lifetime Premium Member
Joined
Nov 13, 2016
Messages
10,293
Reaction score
13,831
Location
Hill Country, TX
Maybe this will help.

One thing I am concerned about is making people trust Theia 100%. It's going to be so quiet that many people will think it's broken just from BSM filtering. If we are going to rely on whitelist 100% then it would need to be PERFECT and we would need to collect every single emitter ever - if someone gets a ticket when Theia didn't alert because we failed to collect a police gun, they would never trust AI again.

At launch it will probably be a mix of white list and black list just because it seems dumb to sit on what is the best detector in the world for another two years while we do data collection - people would still benefit tremendously having range, real camera detection, IO Scout, zero bsms, etc while we continue to make whitelisting perfect. Believe it or not, there are a very limited amount of false alerts to collect since so many emitters are shared across vehicle platforms and then again in safety signs. With blacklisting, it's not a big deal if we miss one by accident - if a customer submits a support ticket we can just go collect it, train, and update. But if we miss an emitter in the whitelist model, someone can get a ticket. I am picturing something like a whitelist with a certain % certainty, and if it's below that it defaults to "unknown KA" or something like that to be safe. I want to build multiple layers of redundancy in it
So for launch, while users will likely have the option of pure whitelist or blacklist, the default setting will really be a safety-net version of both unless and until Radenso feels they have sufficiently covered known threat emitters to the point where they’re comfortable saying the whitelist is complete enough to rely on it exclusively, 100%.

My interpretation of what is likely an oversimplified but generally correct idea of a failsafe logic approach:
Code:
Identify {
    If Whitelist = Recognized
        Alert
    Elseif Blacklist = Recognized
        Ignore
    Else
        Alert
    }
PS They’ve reiterated that they intend to cover international markets as well, not just the US, which means cataloging emitters is and will forever be an ongoing project. Hell, hard-up Mexican Federales have been documented using Bushnells, so...it’s a fun challenge. Right?
 
Last edited:

joe val

PSL +5
Intermediate User
Joined
Aug 31, 2016
Messages
227
Reaction score
121
Video editing is so hard! And time consuming! It's just one more skill that I am forcing myself to try to learn - when you are a small company, you kind of need every employee to know a little bit about everything since we can't afford specialists. I am sure I made tons of mistakes and there is so much more I wish I could do better with, but until the $$$ comes in in from Theia and I can hire a real professional you guys are stuck with my edits! Thank you for the kind words, hopefully we continue to learn what works and what doesn't and always keep improving each week.
Post automatically merged:



It's not even so much trust as the difficulty of getting a clean, unpolluted recoding. Believe it or not, since we are using really advanced DSP we can actually see signals below the noise floor. This is problematic if we let most people record and send us samples, because there is no easy way for the average user to see if a signal below the noise floor is present or not visually. But if it is, it will screw up the AI inferencing.

Basically, we need to be super, super, super careful about the quality of our training data or it all goes to hell.

No need to worry about the quality of editing if Theia is all that it seems to projected to be. After all you are in the RD industry not a editing company. BTW I didn't think the editing was bad at all.
 

Discord Server

Latest threads

Forum statistics

Threads
81,723
Messages
1,240,930
Members
20,704
Latest member
shawn724
Top